net: speedup udp receive path

Since commit 95766fff ([UDP]: Add memory accounting.),
each received packet needs one extra sock_lock()/sock_release() pair.

This added latency because of possible backlog handling. Then later,
ticket spinlocks added yet another latency source in case of DDOS.

This patch introduces lock_sock_bh() and unlock_sock_bh()
synchronization primitives, avoiding one atomic operation and backlog
processing.

skb_free_datagram_locked() uses them instead of full blown
lock_sock()/release_sock(). skb is orphaned inside locked section for
proper socket memory reclaim, and finally freed outside of it.

UDP receive path now take the socket spinlock only once.

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

igb: Clean up left over prototype of igb_get_hw_dev_name()

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

wireless: Fix merge.

in your merge in 5c01d5669356e13f0fb468944c1dd4c6a7e978ad you added "int
i;" into wl1271_main.c which is unused in that function.

This patch fixes the merge problem:

Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Bugfix: Link selection was swapped in switch.

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

caif: Bugfixes in CAIF netdevice for close and flow control

Changes:
o Bugfix: Flow control was causing the device to be destroyed.
o Bugfix: Handle CAIF channel connect failures.
o If the underlying link layer is gone the net-device is no longer removed,
  but closed.

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

caif: Rewritten socket implementation

Changes:
 This is a complete re-write of the socket layer. Making the socket
 implementation more aligned with the other socket layers and using more
 of the support functions available in sock.c. Lots of code is copied
 from af_unix (and some from af_irda).
 Non-blocking mode should be working as well.

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

caif: Disconnect without waiting for response

Changes:
o Function cfcnfg_disconn_adapt_layer is changed to do asynchronous
  disconnect, not waiting for any response from the modem. Due to this
  the function cfcnfg_linkdestroy_rsp does nothing anymore.
o Because disconnect may take down a connection before a connect response
  is received the function cfcnfg_linkup_rsp is checking if the client is
  still waiting for the response, if not a disconnect request is sent to
  the modem.
o cfctrl is no longer keeping track of pending disconnect requests.
o Added function cfctrl_cancel_req, which is used for deleting a pending
  connect request if disconnect is done before connect response is received.
o Removed unused function cfctrl_insert_req2
o Added better handling of connect reject from modem.

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

caif: Add reference counting to service layer

Changes:
o Added functions cfsrvl_get and cfsrvl_put.
o Added support release_client to use by socket and net device.
o Increase reference counting for in-flight packets from cfmuxl

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

caif: Rename functions in cfcnfg and caif_dev

Changes:
 o Renamed cfcnfg_del_adapt_layer to cfcnfg_disconn_adapt_layer
 o Fixed typo cfcfg to cfcnfg
 o Renamed linkid to channel_id
 o Updated documentation in caif_dev.h
 o Minor formatting changes

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

caif: Ldisc add permission check and mem-alloc error check

Changes:
   o Added permission checks for installing. CAP_SYS_ADMIN and
     CAP_SYS_TTY_CONFIG can install the ldisc.
   o Check if allocation of skb was successful.

Signed-off-by: Sjur Braendeland <sjur.brandeland@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Create multiple TX queues

Create a core TX queue and 2 hardware TX queues for each channel.
If separate_tx_channels is set, create equal numbers of RX and TX
channels instead.

Rewrite the channel and queue iteration macros accordingly.
Eliminate efx_channel::used_flags as redundant.

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

sfc: Test only the first pair of TX queues

This makes no immediate difference, but we definitely do not want
to test all TX queues once we allocate a pair of TX queues to each
channel.

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

sfc: Add Siena PHY BIST and cable diagnostic support

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

sfc: Clean up efx_nic::irq_zero_count

There is no need for this to be unsigned long; make it unsigned int.
It does need a line in kernel-doc, so add that.

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

sfc: Add necessary parentheses to macro definitions in net_driver.h

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

sfc: Break NAPI processing after one ring-full of TX completions

Currently TX completions do not count towards the NAPI budget.  This
means a continuous stream of TX completions can cause the polling
function to loop indefinitely with scheduling disabled.  To avoid
this, follow the common practice of reporting the budget spent after
processing one ring-full of TX completions.

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

sfc: Set PERIODIC_NOEVENT flag for MC_CMD_MAC_STATS

When set, an event is not sent whenever periodic MAC statistics are
raised.  This avoids unnecessary wake-ups.

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

sfc: Update MCDI protocol definitions

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

sfc: Enable IPv6 RSS using random key for Toeplitz hash

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

sfc: Read MEM_STAT for SRM_PERR as well as MEM_PERR errors

Parity errors in different blocks of SRAM may set one of two different
interrupt flags.

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

sfc: Log specific message for failure of NVRAM self-test

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

sfc: Extend the legacy interrupt workarounds

Siena has two problems with legacy interrupts:
  1. There is no synchronisation between the ISR read completion,
     and the interrupt deassert message.
  2. A downstream read at the "wrong" moment can return 0, and
     suppress generating the next interrupt.

Falcon should suffer from both of these, and it appears it does.
Enable EFX_WORKAROUND_15783 on Falcon as well.

Also, when we see queues == 0, ensure we always schedule or rearm
every event queue.

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

sfc: Reconfigure the XAUI serdes after an EM reset

Fix a regression introduced in d3245b28ef2a45ec4e115062a38100bd06229289
"sfc: Refactor link configuration".

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

sfc: Stop masking out XGMII faults over reconfigures

The aim of this code was to avoid a spurious XGMII fault over a MAC
reconfigure. It's less relevant now that the PHY reconfigure isn't
called from the MAC reconfigure.

After applying this patch, our link stress test passed 48 hours of
testing without ever resetting the PHY.

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

sfc: Handle serious errors in exactly one interrupt handler

'Fatal' errors set an interrupt flag associated with a specific event
queue; only read the syndrome vector if we see that queue's flag set
(legacy interrupts) or in the interrupt handler for that queue (MSI).

Do not ignore an interrupt if the fatal error flag is set but specific
error flags are all zero.  Even if we don't schedule a reset, we must
respect the queue mask and rearm the appropriate event queues.

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

sfc: Consistently report short MCDI responses as EIO

In some cases failing functions were returning 0 which is obviously wrong.
In other cases they were returning inappropriate error codes.

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

sfc: Ignore parity errors in the other port's SRAM

Siena has a separate SRAM bank for each port.  On single-port boards
these can be merged together, so each port has an interrupt flag for
parity errors in the other port's SRAM.  Currently we do not enable
such merging and should mask this interrupt source.

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

sky2: use the DMA state API instead of the pci equivalents

This replace the PCI DMA state API (include/linux/pci-dma.h) with the
DMA equivalents since the PCI DMA state API will be obsolete.

No functional change.

For further information about the background:

http://marc.info/?l=linux-netdev&m=127037540020276&w=2

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Remove two prefetch()

1) Even on 64bit arches, sizeof(struct sk_buff) < 256
2) No need to prefetch same pointer twice.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Eilon Greenstein <eilong@broadcom.com>
Acked-by: Eliezer Tamir <eliezer@tamir.org.il>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: disable MSI-X by default on certain Cisco adapters

Due to an errata in 82598 parts MSI-X needs to be disabled
in certain ixgbe devices designed to transfer peer-to-peer
traffic on the PCIe bus. This patch sets the default
interrupt type to MSI rather than MSI-X for specific Cisco
ixgbe adapters.

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Acked-by: John Ronciak <john.ronciak@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: multicast_flood cleanup

Move some declarations around to make it clearer which variables
are being used inside loop.

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

bridge: multicast port group RCU fix

The recently introduced bridge mulitcast port group list was only
partially using RCU correctly. It was missing rcu_dereference()
and missing the necessary barrier on deletion.

The code should have used one of the standard list methods (list or hlist)
instead of open coding a RCU based link list.

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

bridge: multicast flood

Fix unsafe usage of RCU. Would never work on Alpha SMP because
of lack of rcu_dereference()

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

bridge: simplify multicast_add_router

By coding slightly differently, there are only two cases
to deal with: add at head and add after previous entry.

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

ixgbe: add registers etc. printout code just before resetting adapters

This patch adds registers (,tx/rx rings' status and so on) printout
code just before resetting adapters. This will be helpful for detecting
the root cause of adapters reset.

Signed-off-by: Taku Izumi <izumi.taku@jp.fujitsu.com>
Signed-off-by: Koki Sanagi <sanagi.koki@jp.fujitsu.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: add registers etc. printout code just before resetting adapters

This patch adds registers (,tx/rx rings' status and so on) printout
code just before resetting adapters. This will be helpful for detecting
the root cause of adapters reset.

Signed-off-by: Taku Izumi <izumi.taku@jp.fujitsu.com>
Signed-off-by: Koki Sanagi <sanagi.koki@jp.fujitsu.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: add registers etc. printout code just before resetting adapters

This patch adds registers (,tx/rx rings' status and so on) printout
code just before resetting adapters. This will be helpful for detecting
the root cause of adapters reset.

Signed-off-by: Taku Izumi <izumi.taku@jp.fujitsu.com>
Signed-off-by: Koki Sanagi <sanagi.koki@jp.fujitsu.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000: Use netdev_<level>, pr_<level> and dev_<level>

This patch is an alternative to similar patch provided by Joe Perches.

Substitute DPRINTK macro for e_<level> that uses netdev_<level> and dev_<level>
similar to e1000e.
- Convert printk to pr_<level> where applicable.
- Use common #define pr_fmt for the driver.
- Use dev_<level> for displaying text in parts of the driver where the interface
  name is not assigned (like e1000_param.c).
- Better align test with the new macros.

CC: Joe Perches <joe@perches.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Revert "bridge: Use hlist_for_each_entry_rcu() in br_multicast_add_router()"

This reverts commit ff65e8275f6c96a5eda57493bd84c4555decf7b3.

As explained by Stephen Hemminger, the traversal doesn't require
RCU handling as we hold a lock.

The list addition et al. calls, on the other hand, do.

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

ixgbevf: use DMA API instead of PCI DMA functions

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: use DMA API instead of PCI DMA functions

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgb: use DMA API instead of PCI DMA functions

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igbvf: use DMA API instead of PCI DMA functions

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: convert igb from using PCI DMA functions to using DMA API functions

This patch makes it so that igb now uses the DMA API functions instead of
the PCI API functions.  To do this the pci_dev pointer that was in the
rings has been replaced with a device pointer, and as a result all
references to [tr]x_ring->pdev have been replaced with [tr]x_ring->dev.

This patch is based of of work originally done by Nicholas Nunley.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: use DMA API instead of PCI DMA functions

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000: use DMA API instead of PCI DMA functions

Signed-off-by: Nicholas Nunley <nicholasx.d.nunley@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: Use hlist_for_each_entry_rcu() in br_multicast_add_router()

Noticed by Michał Mirosław.

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

cxgb4: set skb->rxhash

Implement the ->set_flags ethtool method to control NETIF_F_RXHASH and
set skb->rxhash to the HW calculated hash accordingly.

Follow Eric Dumazet's suggestion and use the hash value raw.

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

net/sb1250: setup the pdevice within the soc code

doing it within the driver does not look good.
And surely isn't how platform devices were meat to be used.

Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/sb1250: remove CONFIG_SIBYTE_STANDALONE

CONFIG_SIBYTE_STANDALONE is gone since v2.6.31-rc1 ("MIPS: Sibyte:
Remove standalone kernel support")
This is a missing piece.

Signed-off-by: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: disallow to use net_assign_generic externally

Now there's no need to use this fuction directly because it's handled by
register_pernet_device. So to make this simple and easy to understand,
make this static to do not tempt potentional users.

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: increase serial number length

Some boards have longer serial numbers in their VPD, up to 24 bytes.

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

cxgb4: parse the VPD instead of relying on a static VPD layout

Some boards' VPDs contain additional keywords or have longer serial numbers,
meaning the keyword locations are variable.  Ditch the static layout and
use the pci_vpd_* family of functions to parse the VPD instead.

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

net: sk_add_backlog() take rmem_alloc into account

Current socket backlog limit is not enough to really stop DDOS attacks,
because user thread spend many time to process a full backlog each
round, and user might crazy spin on socket lock.

We should add backlog size and receive_queue size (aka rmem_alloc) to
pace writers, and let user run without being slow down too much.

Introduce a sk_rcvqueues_full() helper, to avoid taking socket lock in
stress situations.

Under huge stress from a multiqueue/RPS enabled NIC, a single flow udp
receiver can now process ~200.000 pps (instead of ~100 pps before the
patch) on a 8 core machine.

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

net: batch skb dequeueing from softnet input_pkt_queue

batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
contention when RPS is enabled.

Note: in the worst case, the number of packets in a softnet_data may
be double of netdev_max_backlog.

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

net: Make RFS socket operations not be inet specific.

Idea from Eric Dumazet.

As for placement inside of struct sock, I tried to choose a place
that otherwise has a 32-bit hole on 64-bit systems.

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

ixgbe: Properly display 1 gig downshift warning for backplane

Description: When using Intel smartspeed, the patch displays a
warning when the link down shifts to 1 Gig.

Signed-off-by: Anjali Singhai <anjali.singhai@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: cleanup ethtool autoneg input

The way we were setting autoneg via ethtool was inconstant with that
of our other drivers.  It will change the following:

If autoneg is off:
>ethtool -a eth0
Pause parameters for eth0:

Autonegotiate:  off
RX:             off
TX:             off

Before:
>ethtool -A eth0 autoneg on
>ethtool -a eth0
Pause parameters for eth0:

Autonegotiate:  off
RX:             off
TX:             off

Now:
>ethtool -A eth0 autoneg on
>ethtool -a eth0
Pause parameters for eth0:

Autonegotiate:  on
RX:             on
TX:             on

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>

ixgbevf: Fix link speed display

The ixgbevf driver would always report 10Gig speeds even when the link
speed is downshifted to 1Gig.  This patch fixes that problem.

Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: reimplement softnet_data.output_queue as a FIFO queue

reimplement softnet_data.output_queue as a FIFO queue to keep the
fairness among the qdiscs rescheduled.

Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
----
 include/linux/netdevice.h |    1 +
 net/core/dev.c            |   22 ++++++++++++----------
 2 files changed, 13 insertions(+), 10 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of git://git./linux/kernel/git/kaber/ipmr-2.6

ixgb: Use pr_<level> and netdev_<level>

Convert DEBUGOUTx to pr_debug
Convert DEBUGFUNC to more commonly used ENTER
Convert mac address output to %pM
Use #define pr_fmt
Convert a few printks to pr_<level>
Improve ixgb_mc_addr_list_update: use a temporary for current mc address
Use etherdevice.h functions for mac address testing

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: ixgbe_down needs to stop dev_watchdog

There is a small race between when the tx queues are stopped
and when netif_carrier_off() is called in ixgbe_down.  If the
dev_watchdog() timer fires during this time it is possible for
a false tx timeout to occur.

This patch moves the netif_carrier_off() so that it is called before
the tx queues are stopped preventing the dev_watchdog timer from
detecting false tx timeouts.  The race is seen occosionally when
FCoE or DCB settings are being configured or changed.

Testing note, running ifconfig up/down will not reproduce this
issue because dev_open/dev_close call dev_deactivate() and then
dev_activate().

Signed-off-by: John Fastabend <john.r.fastabend@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>

igb: add support for reporting 5GT/s during probe on PCIe Gen2

This change corrects the fact that we were not reporting Gen2 link speeds
when we were in fact connected at Gen2 rates.

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

ixgbe: fix bug when EITR=0 causing no writebacks

writebacks can be held indefinitely by hardware if EITR=0, when
combined with TXDCTL.WTHRESH=8.  When EITR=0, WTHRESH should be
set back to zero.

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

ixgbe: enable extremely low latency

82598/82599 can support EITR == 0, which allows for the
absolutely lowest latency setting in the hardware.  This disables
writeback batching and anything else that relies upon a delayed
interrupt. This patch enables the feature of "override" when a
user sets rx-usecs to zero, the driver will respect that setting
over using RSC, and automatically disable RSC.  If rx-usecs is
used to set the EITR value to 0, then the driver should disable
LRO (aka RSC) internally until EITR is set to non-zero again.

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

igbvf: double increment nr_frags

There is no need to increment nr_frags because skb_fill_page_desc increments
it.

Signed-off-by: Koki Sanagi <sanagi.koki@jp.fujitsu.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: double increment nr_frags

There is no need to increment nr_frags because skb_fill_page_desc increments
it.

Signed-off-by: Koki Sanagi <sanagi.koki@jp.fujitsu.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: fix a lockdep rcu warning in __sk_dst_set()

__sk_dst_set() might be called while no state can be integrated in a
rcu_dereference_check() condition.

So use rcu_dereference_raw() to shutup lockdep warnings (if
CONFIG_PROVE_RCU is set)

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rps: inet_rps_save_rxhash() argument is not const

const qualifier on sock argument is misleading, since we can modify rxhash.

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

TCP: avoid to send keepalive probes if receiving data

RFC 1122 says the following:
...
  Keep-alive packets MUST only be sent when no data or
  acknowledgement packets have been received for the
  connection within an interval.
...

The acknowledgement packet is reseting the keepalive
timer but the data packet isn't. This patch fixes it by
checking the timestamp of the last received data packet
too when the keepalive timer expires.

Signed-off-by: Flavio Leitner <fleitner@redhat.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: multicast router list manipulation

I prefer that the hlist be only accessed through the hlist macro
objects. Explicit twiddling of links (especially with RCU) exposes
the code to future bugs.

Compile tested only.

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

bridge: use is_multicast_ether_addr

Use existing inline function.

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

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

Conflicts:
drivers/net/e100.c
drivers/net/e1000e/netdev.c

net: suppress RCU lockdep false positive in twsk_net()

Calls to twsk_net() are in some cases protected by reference counting
as an alternative to RCU protection.  Cases covered by reference counts
include __inet_twsk_kill(), inet_twsk_free(), inet_twdr_do_twkill_work(),
inet_twdr_twcal_tick(), and tcp_timewait_state_process().  RCU is used
by inet_twsk_purge().  Locking is used by established_get_first()
and established_get_next().  Finally, __inet_twsk_hashdance() is an
initialization case.

It appears to be non-trivial to locate the appropriate locks and
reference counts from within twsk_net(), so used rcu_dereference_raw().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb3: Wait longer for control packets on initialization

In some Power7 platforms, when using VIOS (Virtual I/O Server), we
need to wait longer for control packets to finish transfer during
initialization.
Without this change, initialization may fail prematurely.

Signed-off-by: Wen Xiong <wenxiong@us.ibm.com>
Signed-off-by: Andre Detsch <adetsch@br.ibm.com>
Acked-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: enable/disable ASPM L0s and L1 and ERT according to hardware errata

Prompted by a previous patch submitted by Matthew Garret <mjg@redhat.com>,
further digging into errata documentation reveals the current enabling or
disabling of ASPM L0s and L1 states for certain parts supported by this
driver are incorrect.  82571 and 82572 should always disable L1.  For
standard frames, 82573/82574/82583 can enable L1 but L0s must be disabled,
and for jumbo frames 82573/82574 must disable L1.  This allows for some
parts to enable L1 in certain configurations leading to better power
savings.

Also according to the same errata, Early Receive (ERT) should be disabled
on 82573 when using jumbo frames.

Cc: Matthew Garret <mjg@redhat.com>
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>

ixgbe: Power down PHY during driver resets

The PHY laser is still on during driver init.  It's allowing
garbage to hit our FIFO, which eventually can cause the entire
device to die.  Power down the laser while setting up the device,
and re-enable the laser before getting link.

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

bridge: Fix build of ipv6 multicast code.

Based upon a report from Stephen Rothwell:

--------------------
net/bridge/br_multicast.c: In function 'br_ip6_multicast_alloc_query':
net/bridge/br_multicast.c:469: error: implicit declaration of function 'csum_ipv6_magic'

Introduced by commit 08b202b6726459626c73ecfa08fcdc8c3efc76c2 ("bridge
br_multicast: IPv6 MLD support") from the net tree.

csum_ipv6_magic is declared in net/ip6_checksum.h ...
--------------------

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

r8169: more broken register writes workaround

78f1cd02457252e1ffbc6caa44a17424a45286b8 ("fix broken register writes")
does not work for Al Viro's r8169 (XID 18000000).

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

r8169: failure to enable mwi should not be fatal

Few (6) network drivers enable mwi explicitly. Fewer worry about a
failure.

It is not a fix but it should avoid some annoyance like
http://bugzilla.kernel.org/show_bug.cgi?id=15454

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Cc: Conrad Kostecki <conikost@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge br_multicast: Ensure to initialize BR_INPUT_SKB_CB(skb)->mrouters_only.

Even with commit 32dec5dd0233ebffa9cae25ce7ba6daeb7df4467 ("bridge
br_multicast: Don't refer to BR_INPUT_SKB_CB(skb)->mrouters_only
without IGMP snooping."), BR_INPUT_SKB_CB(skb)->mrouters_only is
not appropriately initialized if IGMP/MLD snooping support is
compiled and disabled, so we can see garbage.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge br_multicast: Ensure to initialize BR_INPUT_SKB_CB(skb)->mrouters_only.

Even with commit 32dec5dd0233ebffa9cae25ce7ba6daeb7df4467 ("bridge
br_multicast: Don't refer to BR_INPUT_SKB_CB(skb)->mrouters_only
without IGMP snooping."), BR_INPUT_SKB_CB(skb)->mrouters_only is
not appropriately initialized if IGMP snooping support is
compiled and disabled, so we can see garbage.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ieee802154: Fix oops during ieee802154_sock_ioctl

Trying to run izlisten (from lowpan-tools tests) on a device that does not
exists I got the oops below. The problem is that we are using get_dev_by_name
without checking if we really get a device back. We don't in this case and
writing to dev->type generates this oops.

[Oops code removed by Dmitry Eremin-Solenikov]

If possible this patch should be applied to the current -rc fixes branch.

Signed-off-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: Dmitry Eremin-Solenikov <dbaryshkov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: use sk_sleep()

Commit aa395145 (net: sk_sleep() helper) missed three files in the
conversion.

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

pppoe: use pppoe_pernet instead of directly net_generic

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

phonet: use phonet_pernet instead of directly net_generic

As in for example pppoe introduce phonet_pernet and use it instead of calling
net_generic directly.

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Fix INTx fallback when MSI fails

tg3: Fix INTx fallback when MSI fails

MSI setup changes the value of irq_vec in struct tg3 *tp.
This attribute must be taken into account and restored before
we try to do a new request_irq for INTx fallback.

In powerpc, the original code was leading to an EINVAL return within
request_irq, because the driver was trying to use the disabled MSI
virtual irq number instead of tp->pdev->irq.

Signed-off-by: Andre Detsch <adetsch@br.ibm.com>
Acked-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ipmr: add support for dumping routing tables over netlink

The ipmr /proc interface (ip_mr_cache) can't be extended to dump routes
from any tables but the main table in a backwards compatible fashion since
the output format ends in a variable amount of output interfaces.

Introduce a new netlink interface to dump multicast routes from all tables,
similar to the netlink interface for regular routes.

Signed-off-by: Patrick McHardy <kaber@trash.net>

net: rtnetlink: decouple rtnetlink address families from real address families

Decouple rtnetlink address families from real address families in socket.h to
be able to add rtnetlink interfaces to code that is not a real address family
without increasing AF_MAX/NPROTO.

This will be used to add support for multicast route dumping from all tables
as the proc interface can't be extended to support anything but the main table
without breaking compatibility.

This partialy undoes the patch to introduce independant families for routing
rules and converts ipmr routing rules to a new rtnetlink family. Similar to
that patch, values up to 127 are reserved for real address families, values
above that may be used arbitrarily.

Signed-off-by: Patrick McHardy <kaber@trash.net>

net: fib_rules: mark arguments to fib_rules_register const and __net_initdata

fib_rules_register() duplicates the template passed to it without modification,
mark the argument as const. Additionally the templates are only needed when
instantiating a new namespace, so mark them as __net_initdata, which means
they can be discarded when CONFIG_NET_NS=n.

Signed-off-by: Patrick McHardy <kaber@trash.net>

ipv6: Fix inet6_csk_bind_conflict()

Commit fda48a0d7a84 (tcp: bind() fix when many ports are bound)
introduced a bug on IPV6 part.
We should not call ipv6_addr_any(inet6_rcv_saddr(sk2)) but
ipv6_addr_any(inet6_rcv_saddr(sk)) because sk2 can be IPV4, while sk is
IPV6.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netns: rename unregister_pernet_subsys parameter

Stay consistent with other functions and with comment also and name
pernet_operations parameter properly.

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rps: optimize rps_get_cpu()

optimize rps_get_cpu().

don't initialize ports when we can get the ports. one memory access
for ports than two.

Signed-off-by: Changli Gao <xiaosuo@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e100: Fix the TX workqueue race

Nothing stops the workqueue being left to run in parallel with close or a
few other operations. This causes double unmaps and the like.

See kerneloops.org #1041230 for an example

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sky2: add support for receive hashing

Sky2 hardware supports hardware receive hash calculation.
Now that Receive Packet Steering is available, add support
to enable it.

This version does not depend on CONFIG_RPS. Also set_flags rejects
all values except RXHASH, so driver won't have to change next time
somebody adds a new one.

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

Merge branch 'net-next-2.6_20100423a/br/br_multicast_v3' of git://git.linux-ipv6.org/gitroot/yoshfuji/linux-2.6-next

IPv6: Complete IPV6_DONTFRAG support

Finally add support to detect a local IPV6_DONTFRAG event
and return the relevant data to the user if they've enabled
IPV6_RECVPATHMTU on the socket.  The next recvmsg() will
return no data, but have an IPV6_PATHMTU as ancillary data.

Signed-off-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6: Add dontfrag argument to relevant functions

Add dontfrag argument to relevant functions for
IPV6_DONTFRAG support, as well as allowing the value
to be passed-in via ancillary cmsg data.

Signed-off-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6: data structure changes for new socket options

Add underlying data structure changes and basic setsockopt()
and getsockopt() support for IPV6_RECVPATHMTU, IPV6_PATHMTU,
and IPV6_DONTFRAG.  IPV6_PATHMTU is actually fully functional
at this point.

Signed-off-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>