gianfar: Add Multiple Queue Support

This patch introduces multiple Tx and Rx queues.
The incoming packets can be classified into different queues
based on filer rules (out of scope of this patch). The number
of queues enabled will be based on a DTS entries fsl,num_tx_queues
and fsl,num_rx_queues.

Although we are enabling multiple queues, the interrupt coalescing
is on per device level (etsec-1.7 doesn't support multiple rxics
and txics).

Signed-off-by: Sandeep Gopalpet <Sandeep.Kumar@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gianfar: Introduce logical group support.

This patch introduces the group structure. The elements of this
structure are the interrupt lines, their corresponding names,
the register memory map.
The elements for this group are factored out from the gfar_private
structure. The introduction of group structure will help in
providing support for newer versions of etsec.

Currently, the support is present only for single group and
single tx/rx queues.

Signed-off-by: Sandeep Gopalpet <Sandeep.Kumar@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gianfar: Add per queue structure support

This patch introduces per tx and per rx queue structures.
Earlier the members of these structures were inside the
gfar_private structure.

Moving forward if we want to support multiple queues, we need
to refactor the gfar_private structure so that introduction of
multiple queues is easier.

Signed-off-by: Sandeep Gopalpet <Sandeep.Kumar@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Update version to 3.103

This patch updates the tg3 version to 3.103.

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 / broadcom: Optionally disable TXC if no link

This patch adds code to disable the TXC and RXC reference clocks if link
is not available.

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 / broadcom: Add APD support for GPHYs

This patch adds an RXC auto power-down feature to the code that supports
the gphys.

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 / broadcom: Add code to disable rxc refclk

The 5785 does not use the RXC reference clock.  Turning it off is
desirable as it saves power.

By default, the 50610 enables the RXC reference clock and the 50610M
disables it.  Presumably this is one of the reasons why the hardware
architect chose one over the other.

Adding a "rx reference clock disable" flag is not the ideal way to
describe the option, as it would force the MAC using a 50610M to set
the flag.  Ideally we want the flags to represent opt-in behavior that
deviates from hardware defaults.  Furthermore, the lack of a
"disable" flag implies that the requester wants the rx reference clock
enabled, which doesn't necessarily follow.

By presenting the option as a passive statement (rx reference clock
unused) rather than a command, I hope to convey an opt-in option to
disable the rx reference clock that falls back to hardware defaults if
not set.  A secondary benefit of this is that it keeps the
intelligence about phy defaults in the broadcom module where it belongs
and allows the broadcom module more latitude should a bug arise.

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 / broadcom: Refine AC131 APD support

Auto power-down (APD) support is a power-saving feature.  It should be
selectively enabled since it might expose MAC bugs.  This patch changes
the code to enable APD only if the PHY_BRCM_AUTO_PWRDWN_ENABLE flag is
set.  The tg3 driver was changed to set this bit.

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 / broadcom: Add PHY_BRCM_CLEAR_RGMII_MODE flag

Broadcom 50610M parts changed the default definitions of the RGMII mode
shadow register.  The 5785 needs the RGMII mode selection bits [4:3]
cleared.

The default value of the remaining bits in this register are zero.
Rather than unnecessarily burn an extra bit in the dev_flags member in
an attempt to enumerate all possible combinations, this patch take a
more course grained approach and labels the option as "clear all bits".

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

broadcom: Consolidate dev_flags definitions

This patch moves all the dev_flags enumerations outside the broadcom.c
file to include/linux/brcmphy.h.  The existing flags were not used yet
and have been re-enumerated to avoid conflicts.

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 50610M phy ID for 5785

This patch adds the 50610M phy ID for 5785.

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

broadcom: Fix slow link problem

When a 50610 or 50610M is paired against particular remote partners,
link is slow to come up.  This patch works around the problem.

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

broadcom: Isolate phy dsp accesses

This patch consolidates the code that requires the SMDSP clock to be
enabled into a single function that (hopefully) makes the dependency
obvious.

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: Extend loopback test timeout

This patch extends the loopback test timeout from 250 usec to 350 usec.
When the 5785 is paired against an AC131 phy, the older timeout is
just a little too close to the expected performance based on timings.

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: 5785: Set port mode to MII when link down

This patch sets the port mode to MII when the link is down for the 5785.
Setting the port mode to MII instead of GMII saves power.

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 AC131 power down support

The AC131 does not respect the power down bit (bit 11) of the MII
Control Register (reg 0x0).  Instead, software is required to put the
phy into standby power down mode through the shadow register set.  This
patch implements support for the AC131 standby power down mode.

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: Improve 5785 PCIe performance

This patch improves 5785 performance by allowing the write DMA engine to
request larger DMA burst sizes than it otherwise would.

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 more PCI DMA map error checking

This patch adds code to check the status of pci_map_single() before
allowing rx buffers to be used.  It also converts the pci_map_single()
call in tg3_run_loopback() to use skb_dma_map() instead.

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

tg3: Prevent a PCIe tx glitch

This patch prevents a PCIe tx glitch by allowing the transmitter to go
to a low power state.

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: Convert PHY_ADDR => TG3_PHY_MII_ADDR

This patch renames the PHY_ADDR preprocessor definition.  The following
patch will identify a new member on the MDIO bus, so we want this
preprocessor definition to be a little more descriptive.

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 disappearing 57780 devices

Under certain power saving conditions, 57780 asic rev devices might
disappear from the system.  The fix is to disallow the PCIe PLL from
powering down.

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 5906 transmit hangs

The 5906 has trouble with fragments that are less than 8 bytes in size.
This patch works around the problem by pivoting the 5906's transmit
routine to tg3_start_xmit_dma_bug() and introducing a new SHORT_DMA_BUG
flag that enables code to detect and react to the problematic condition.

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

tg3: Assign flags to fixes in start_xmit_dma_bug

This patch adds a flag for each bug workaround in
tg3_start_xmit_dma_bug().  This is prep work for the following patch.

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

qlge: Change naming on vlan API.

Change name on vlan_rx_add, kill, register to match other driver API.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlge: Add ethtool self-test.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: no more dev_put() in datagram_send_ctl()

Avoids touching device refcount in datagram_send_ctl(), thanks to RCU

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

ipv6: no more dev_put() in inet6_bind()

Avoids touching device refcount in inet6_bind(), thanks to RCU

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

ip6tnl: less dev_put() calls

Using dev_get_by_index_rcu() in ip6_tnl_rcv_ctl() & ip6_tnl_xmit_ctl()
avoids touching device refcount.

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

packet: less dev_put() calls

- packet_sendmsg_spkt() can use dev_get_by_name_rcu() to avoid touching device refcount.

- packet_getname_spkt() & packet_getname() can use dev_get_by_index_rcu() to
  avoid touching device refcount too.

tpacket_snd() & packet_snd() can not use RCU yet because they can sleep when
allocating skb.

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

pppoe: RCU locking in get_item_by_addr()

Use dev_get_by_name_rcu() instead of dev_get_by_name(),
to avoid touching device refcount in hotpath.

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

ifb: RCU locking avoids touching dev refcount

Avoids touching dev refcount in hotpath

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

net: RCU locking for simple ioctl()

All ioctls() implemented by dev_ifsioc_locked() :
SIOCGIFFLAGS, SIOCGIFMETRIC, SIOCGIFMTU, SIOCGIFHWADDR,
SIOCGIFSLAVE, SIOCGIFMAP, SIOCGIFINDEX & SIOCGIFTXQLEN
can use RCU lock instead of dev_base_lock rwlock

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

icmp: icmp_send() can avoid a dev_put()

We can avoid touching device refcount in icmp_send(),
using dev_get_by_index_rcu()

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

ipv4: inetdev_by_index() switch to RCU

Use dev_get_by_index_rcu() instead of __dev_get_by_index() and
dev_base_lock rwlock

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

veth: Fix unregister_netdevice_queue for veth

I tested the recent unregister many changes and got a weird,
nasty and seemingly unrelasted kernel oops. Changing
unregister_netdevice_queue to use list_move_tail fixes
the problem for me.

ip link add type veth
rmmod veth

ls /sys/class/net/
showed one of the veth devices still present.

A subsequent ip link oopsed the box.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Introduce dev_get_by_name_rcu()

Some workloads hit dev_base_lock rwlock pretty hard.
We can use RCU lookups to avoid touching this rwlock
(and avoid touching netdevice refcount)

netdevices are already freed after a RCU grace period, so this patch
adds no penalty at device dismantle time.

However, it adds a synchronize_rcu() call in dev_change_name()

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

NET:KS8695: add API for get rx interrupt bit

1. Add API Add k8695_get_rx_enable_bit() for get Rx interrupt
enable/status bit.
2. add some comment or document about some functions and variables.
3.  update driver version to "1.02"
4. add napi_enable() and napi_disable() in open/close file method.

Signed-off-by: Figo.zhang <figo1802@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RDS/IB+IW: Move recv processing to a tasklet

Move receive processing from event handler to a tasklet.
This should help prevent hangcheck timer from going off
when RDS is under heavy load.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RDS: Do not send congestion updates to loopback connections

This issue was discovered by HP's Pradeep and fixed in OFED
1.3, but not fixed in later versions, since the fix's implementation
was not immediately applyable to the later code. This patch should
do the trick for 1.4+ codebases.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RDS: Fix panic on unload

Remove explicit destruction of passive connection when destroying
active end of the connection. The passive end is also on the
device's connection list, and will thus be cleaned up properly.
Panic was caused by trying to clean it up twice.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RDS: Fix potential race around rds_i[bw]_allocation

"At rds_ib_recv_refill_one(), it first executes atomic_read(&rds_ib_allocation)
for if-condition checking,

and then executes atomic_inc(&rds_ib_allocation) if the condition was
not satisfied.

However, if any other code which updates rds_ib_allocation executes
between these two atomic operation executions,
it seems that it may result race condition. (especially when
rds_ib_allocation + 1 == rds_ib_sysctl_max_recv_allocation)"

This patch fixes this by using atomic_inc_unless to eliminate the
possibility of allocating more than rds_ib_sysctl_max_recv_allocation
and then decrementing the count if the allocation fails. It also
makes an identical change to the iwarp transport.

Reported-by: Shin Hong <hongshin@gmail.com>
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RDS: Add GET_MR_FOR_DEST sockopt

RDS currently supports a GET_MR sockopt to establish a
memory region (MR) for a chunk of memory. However, the fastreg
method ties a MR to a particular destination. The GET_MR_FOR_DEST
sockopt allows the remote machine to be specified, and thus
support for fastreg (aka FRWRs).

Note that this patch does *not* do all of this - it simply
implements the new sockopt in terms of the old one, so applications
can begin to use the new sockopt in preparation for cutover to
FRWRs.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bond: Get the rtnl_link_ops support correct

- Don't call rtnl_link_unregister if rtnl_link_register fails
- Set .priv_size so we aren't stomping on uninitialized memory
  when we use netdev_priv, on bond devices created with
  ip link add type bond.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bond: Add support for multiple network namespaces

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bond: Implement a basic set of rtnl link ops

This implements a basic set of rtnl link ops and takes advantage of
the fact that rtnl_link_unregister kills all of the surviving
devices to all us to kill bond_free_all.  A module alias
is added so ip link add can pull in the bonding module.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bond: Simplify bond device destruction

Manually inline the code from bond_deinit to bond_uninit.  bond_uninit
is the only caller and it is short.

Move the call of bond_release_all from the netdev notifier into
bond_uninit.  The call site is effectively the same and performing
the call explicitly allows all the paths for destroying a
bonding device to behave the same way.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bond: Simplify bond_create.

Stop calling dev_get_by_name to see if the bond device already
exists.  register_netdevice already does that.

Stop calling bond_deinit if register_netdevice fails as bond_uninit
is guaranteed to be called if bond_init succeeds.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bond: Simply bond sysfs group creation

This patch delegates the work of creating the sysfs groups
to the netdev layer and ultimately to the device layer.  This
closes races between uevents.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Allow devices to specify a device specific sysfs group.

This isn't beautifully abstracted, but it is simple,
simplifies uses and so far is only needed for the bonding driver.

Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: use hlist_for_each_entry()

Small cleanup of __dev_get_by_name() and __dev_get_by_index()
to use hlist_for_each_entry() : They'll look like their _rcu variant.

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

veth: Fix veth_dellink method

In commit 23289a37e2b127dfc4de1313fba15bb4c9f0cd5b
(net: add a list_head parameter to dellink() method),
I forgot to actually use this parameter in veth_dellink.

I remember feeling a bit uncomfortable about veth_close(),
because it does :

netif_carrier_off(dev);
netif_carrier_off(priv->peer);

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

vlan: cleanup multiple unregistrations

The temporary copy of the VLAN group is not neccessary since the lower device
is already in the process of being unregistered, if it was neccessary the
memset of the global group would introduce a race condition.

With this removed, the changes to the original code are only a few lines, so
remove the new function and move the code back into vlan_device_event().

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

convert kaweth to use usb_reset_configuration()

For USB 3.0 it is necessary that all drivers use the standard
API to reset a configuration. This removes a home-grown
implementation.

Signed-off-by: Oliver Neukum <oliver@neukum.org>
Hi David,

please take this for the next merge window.

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

ax25: unsigned cannot be less than 0 in ax25_ctl_ioctl()

struct ax25_ctl_struct member `arg' is unsigned and cannot be less
than 0.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sky2: version 1.26

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

sky2: 88E8059 support

Tentative support for newer Marvell hardware including
the Yukon-2 Optima chip. Do not have hatdware to test this yet,
code is based on vendor driver.

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

sky2: workarounds for Yukon-2 supreme

Changes related to support of Yukon supreme chip.
Don't have this chip version to test on,
these are reverse engineered from the vendor (GPL) driver.

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

sky2: fix receive pause thresholds

Program the receive pause thresholds differently depending on
chip version. This cloned from from the vendor (GPL) driver.

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

sky2: add register definitions for new chips

This adds infrastructure for the newer chip versions and workarounds.
Extracted from the vendor (GPL) driver.

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

sky2: add SK-9E21M device id

This is a new ID that just showed up in latest vendor driver.

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

sfc: Enable heuristic selection between page and skb RX buffers

Now that we can tell whether GRO is being applied, this heuristic is
effective once more.

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

sfc: Feed GRO result into RX allocation policy and interrupt moderation

When GRO is successfully merging received packets, we should allocate
raw page buffers rather than skbs that will be discarded by GRO.
Otherwise, we should allocate skbs.

GRO also benefits from higher interrupt moderation, so increase the
score for mergeable RX packets.

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

gro: Change all receive functions to return GRO result codes

This will allow drivers to adjust their receive path dynamically
based on whether GRO is being applied successfully.

Currently all in-tree callers ignore the return values of these
functions and do not need to be changed.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

gro: Name the GRO result enumeration type

This clarifies which return and parameter types are GRO result codes
and not RX result codes.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

net: Fix 'Re: PACKET_TX_RING: packet size is too long'

Currently PACKET_TX_RING forces certain amount of every frame to remain
unused. This probably originates from an early version of the
PACKET_TX_RING patch that in fact used the extra space when the (since
removed) CONFIG_PACKET_MMAP_ZERO_COPY option was enabled. The current
code does not make any use of this extra space.

This patch removes the extra space reservation and lets userspace make
use of the full frame size.

Signed-off-by: Gabor Gombas <gombasg@sztaki.hu>
Signed-off-by: David S. Miller <davem@davemloft.net>

net,socket: introduce DECLARE_SOCKADDR helper to catch overflow at build time

proto_ops->getname implies copying protocol specific data
into storage unit (particulary to __kernel_sockaddr_storage).
So when we implement new protocol support we should keep such
a detail in mind (which is easy to forget about).

Lets introduce DECLARE_SOCKADDR helper which check if
storage unit is not overfowed at build time.

Eventually inet_getname is switched to use DECLARE_SOCKADDR
(to show example of usage).

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

net: Introduce dev_get_by_index_rcu()

Some workloads hit dev_base_lock rwlock pretty hard.
We can use RCU lookups to avoid touching this rwlock.

netdevices are already freed after a RCU grace period, so this patch
adds no penalty at device dismantle time.

dev_ifname() converted to dev_get_by_index_rcu()

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

atm: Cleanup redundant tests on unsigned

The variables are unsigned so the `< 0' test always fails, the
other part of the test catches wrapped values.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Cleanup redundant tests on unsigned

optlen is unsigned so the `< 0' test is never true.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Cleanup redundant tests on unsigned

If there is data, the unsigned skb->len is greater than 0.

rt.sigdigits is unsigned as well, so the test `>= 0' is
always true, the other part of the test catches wrapped
values.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gigaset: convert strcmp chain to table lookup

Replace the sequence of strcmp calls for interpreting ZSAU parameter
strings by a table of known strings and lookup loop to improve
readability.

Impact: readability improvement, no functional change
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

gigaset: checkpatch cleanup of ev-layer.c

On more step towards the holy grail of checkpatch.pl silence.

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

bas_gigaset: checkpatch cleanup

On the quest for the holy grail of checkpatch.pl silence.

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

ser_gigaset: checkpatch cleanup

Duly uglified as demanded by checkpatch.pl.

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

gigaset: checkpatch cleanup

Dum sanctis checkpatch.pl'ae legibus obsequimur.

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

usb_gigaset: code cleanup

Reorganize the code of the Gigaset M10x driver to make it more
readable, less redundant, better aligned to the style of other
parts of the driver, and cause fewer checkpatch.pl complaints.

Impact: code reorganization, no functional change
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

gigaset: fix bad assumptions about CAPI skbuffs

The CAPI interface incorrectly assumed that CAPI messages would always
start at the beginning of the data buffer: fix by treating DATA_B3
messages as the link layer header to their payload data. This fix
changes the way acknowledgement information is propagated through the
hardware specific modules and thereby impacts the ISDN4Linux variant
of the driver, too.

Also some assumptions about methods not being called from interrupt
context turned out to be unwarranted; fix by using dev_kfree_skb_any()
wherever non-interrupt context isn't guaranteed.

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

gigaset: fix format string typo in CAPI dial command

A missing dot lead to garbage characters being included in the
dial command generated from a CAPI CONNECT_REQ message, which
interestingly enough worked anyway, illustrating the resilience
of the device.

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

gigaset: CAPI module readability improvements

Replace the "ignoring Additional Info" warning message by better
readable ones citing the specific subparameters being ignored.
Make parts of the code more readable by using a local cmsg
pointer variable.

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

Allow disabling of DSACK TCP option per route

Add and use no DSCAK bit in the features field.

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Sigend-off-by: Ori Finkelman <ori@comsleep.com>
Sigend-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Allow to turn off TCP window scale opt per route

Add and use no window scale bit in the features field.

Note that this is not the same as setting a window scale of 0
as would happen with window limit on route.

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Sigend-off-by: Ori Finkelman <ori@comsleep.com>
Sigend-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Allow disabling TCP timestamp options per route

Implement querying and acting upon the no timestamp bit in the feature
field.

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Sigend-off-by: Ori Finkelman <ori@comsleep.com>
Sigend-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add the no SACK route option feature

Implement querying and acting upon the no sack bit in the features
field.

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Sigend-off-by: Ori Finkelman <ori@comsleep.com>
Sigend-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add dst_feature to query route entry features

Adding an accessor to existing  dst_entry feautres field and
refactor the only supported feature (allfrag) to use it.

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Sigend-off-by: Ori Finkelman <ori@comsleep.com>
Sigend-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Allow tcp_parse_options to consult dst entry

We need tcp_parse_options to be aware of dst_entry to
take into account per dst_entry TCP options settings

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Sigend-off-by: Ori Finkelman <ori@comsleep.com>
Sigend-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Only parse time stamp TCP option in time wait sock

Since we only use tcp_parse_options here to check for the exietence
of TCP timestamp option in the header, it is better to call with
the "established" flag on.

Signed-off-by: Gilad Ben-Yossef <gilad@codefidence.com>
Signed-off-by: Ori Finkelman <ori@comsleep.com>
Signed-off-by: Yony Amit <yony@comsleep.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netdev: usb: dm9601.c can drive a device not supported yet, add support for it

I found that the current version of drivers/net/usb/dm9601.c can be used to
successfully drive a low-power, low-cost network adapter with USB ID
0a46:9000, based on a DM9000E chipset. As no device with this ID is yet
present in the kernel, I have created a patch that adds support for the device
to the dm9601 driver.

Created and tested against linux-2.6.32-rc5.

Signed-off-by: Janusz Krzysztofik <jkrzyszt@tis.icnet.pl>
Acked-by: Peter Korsgaard <jacmet@sunsite.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlge: Fix firmware mailbox command timeout.

The mailbox command process would only process a maximum of 5 unrelated
firmware events while waiting for it's command completion status.
It should process an unlimited number of events while waiting for a maximum of 5 seconds.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlge: Fix EEH handling.

Clean up driver resources without touch the hardware. Add pci
save/restore state.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6mr: Optimize multiple unregistration

Speedup module unloading by factorizing synchronize_rcu() calls

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

ipv6 sit: Optimize multiple unregistration

Speedup module unloading by factorizing synchronize_rcu() calls

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

ipmr: Optimize multiple unregistration

Speedup module unloading by factorizing synchronize_rcu() calls

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

ip6tnl: Optimize multiple unregistration

Speedup module unloading by factorizing synchronize_rcu() calls

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

bridge: Optimize multiple unregistration

Speedup module unloading by factorizing synchronize_rcu() calls

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

be2net: Add the new PCI IDs to PCI_DEVICE_TABLE.

This patch adds the PCI IDs for the next generation chip to the
PCI_DEVICE_ID table.

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

AF_RAW: Augment raw_send_hdrinc to expand skb to fit iphdr->ihl (v2)

Augment raw_send_hdrinc to correct for incorrect ip header length values

A series of oopses was reported to me recently.  Apparently when using AF_RAW
sockets to send data to peers that were reachable via ipsec encapsulation,
people could panic or BUG halt their systems.

I've tracked the problem down to user space sending an invalid ip header over an
AF_RAW socket with IP_HDRINCL set to 1.

Basically what happens is that userspace sends down an ip frame that includes
only the header (no data), but sets the ip header ihl value to a large number,
one that is larger than the total amount of data passed to the sendmsg call.  In
raw_send_hdrincl, we allocate an skb based on the size of the data in the msghdr
that was passed in, but assume the data is all valid.  Later during ipsec
encapsulation, xfrm4_tranport_output moves the entire frame back in the skbuff
to provide headroom for the ipsec headers.  During this operation, the
skb->transport_header is repointed to a spot computed by
skb->network_header + the ip header length (ihl).  Since so little data was
passed in relative to the value of ihl provided by the raw socket, we point
transport header to an unknown location, resulting in various crashes.

This fix for this is pretty straightforward, simply validate the value of of
iph->ihl when sending over a raw socket.  If (iph->ihl*4U) > user data buffer
size, drop the frame and return -EINVAL.  I just confirmed this fixes the
reported crashes.

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

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

e1000e: flow control doesn't re-enable

When changing flow control (pause) parameters, the flow control thresholds
(i.e. when to send XON/XOFF frames) may not be setup correctly on parts
with copper media.  Call the existing e1000_set_fc_watermarks()
function to set these thresholds.

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