net: percpu net_device refcount

We tried very hard to remove all possible dev_hold()/dev_put() pairs in
network stack, using RCU conversions.

There is still an unavoidable device refcount change for every dst we
create/destroy, and this can slow down some workloads (routers or some
app servers, mmap af_packet)

We can switch to a percpu refcount implementation, now dynamic per_cpu
infrastructure is mature. On a 64 cpus machine, this consumes 256 bytes
per device.

On x86, dev_hold(dev) code :

before
        lock    incl 0x280(%ebx)
after:
        movl    0x260(%ebx),%eax
        incl    fs:(%eax)

Stress bench :

(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE)

Before:

real    1m1.662s
user    0m14.373s
sys     12m55.960s

After:

real    0m51.179s
user    0m15.329s
sys     10m15.942s

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

bnx2x: Fixing a typo: added a missing RSS enablement

Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Vladislav Zolotarov <vladz@broadcom.com>
Tested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'dccp' of git://eden-feed.erg.abdn.ac.uk/net-next-2.6

dccp: cosmetics - warning format

This  omits the redundant "DCCP:" in warning messages, since DCCP_WARN() already
echoes the function name, avoiding messages like

   kernel: [10988.766503] dccp_close: DCCP: ABORT -- 209 bytes unread

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

dccp: schedule an Ack when receiving timestamps

This schedules an Ack when receiving a timestamp, exploiting the
existing inet_csk_schedule_ack() function, saving one case in the
`dccp_ack_pending()' function.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

dccp: generalise data-loss condition

This patch generalises the task of determining data loss from RFC 4340, 7.7.1.

Let S_A, S_B be sequence numbers such that S_B is "after" S_A, and let
N_B be the NDP count of packet S_B. Then, using modulo-2^48 arithmetic,
 D = S_B - S_A - 1  is an upper bound of the number of lost data packets,
 D - N_B            is an approximation of the number of lost data packets
                    (there are cases where this is not exact).

The patch implements this as
 dccp_loss_count(S_A, S_B, N_B) := max(S_B - S_A - 1 - N_B, 0)

Signed-off-by: Ivo Calado <ivocalado@embedded.ufcg.edu.br>
Signed-off-by: Erivaldo Xavier <desadoc@gmail.com>
Signed-off-by: Leandro Sales <leandroal@gmail.com>
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

dccp: remove unused argument in CCID tx function

This removes the argument `more' from ccid_hc_tx_packet_sent, since it was
nowhere used in the entire code.

(Btw, this argument was not even used in the original KAME code where the
 function initially came from; compare the variable moreToSend in the
 freebsd61-dccp-kame-28.08.2006.patch kept by Emmanuel Lochin.)

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

dccp: merge now-reduced connect_init() function

After moving the assignment of GAR/ISS from dccp_connect_init() to
dccp_transmit_skb(), the former function becomes very small, so that
a merger with dccp_connect() suggests itself.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

dccp: fix the adjustments to AWL and SWL

This fixes a problem and a potential loophole with regard to seqno/ackno
validity: currently the initial adjustments to AWL/SWL are only performed
once at the begin of the connection, during the handshake.

Since the Sequence Window feature is always greater than Wmin=32 (7.5.2),
it is however necessary to perform these adjustments at least for the first
W/W' (variables as per 7.5.1) packets in the lifetime of a connection.

This requirement is complicated by the fact that W/W' can change at any time
during the lifetime of a connection.

Therefore it is better to perform that safety check each time SWL/AWL are
updated, as implemented by the patch.

A second problem solved by this patch is that the remote/local Sequence Window
feature values (which set the bounds for AWL/SWL/SWH) are undefined until the
feature negotiation has completed.

During the initial handshake we have more stringent sequence number protection;
the changes added by this patch effect that {A,S}W{L,H} are within the correct
bounds at the instant that feature negotiation completes (since the SeqWin
feature activation handlers call dccp_update_gsr/gss()).

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

bnx2: Enable AER on PCIE devices only

To prevent unnecessary error message.  pci_save_state() is also moved to
the end of ->probe() so that all PCI config, including AER state, will be
saved.

Update version to 2.0.18.

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

bnx2: Update firmware to 6.0.x.

- Improved flow control and simplified interface
- Use hardware RSS indirection table instead of the slower firmware-
  based table
- Lower latency interrupt on 5709

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

neigh: reorder struct neighbour fields

Le mardi 12 octobre 2010 à 00:02 +0200, Eric Dumazet a écrit :
> Here is the followup patch.
>
> Thanks !
>

Oops, this was an old version, the up2date ones also took care of "used"
field.

I guess its time for a sleep, sorry again.

[PATCH net-next V2] neigh: reorder struct neighbour fields

(refcnt) and (ha_lock, ha, used, dev, output, ops, primary_key) should
be placed on a separate cache lines.

refcnt can be often written, while other fields are mostly read.

This gave me good result on stress test :

before:

real    0m45.570s
user    0m15.525s
sys     9m56.669s

After:

real    0m41.841s
user    0m15.261s
sys     8m45.949s

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

net dst: use a percpu_counter to track entries

struct dst_ops tracks number of allocated dst in an atomic_t field,
subject to high cache line contention in stress workload.

Switch to a percpu_counter, to reduce number of time we need to dirty a
central location. Place it on a separate cache line to avoid dirtying
read only fields.

Stress test :

(Sending 160.000.000 UDP frames,
IP route cache disabled, dual E5540 @2.53GHz,
32bit kernel, FIB_TRIE, SLUB/NUMA)

Before:

real    0m51.179s
user    0m15.329s
sys     10m15.942s

After:

real 0m45.570s
user 0m15.525s
sys 9m56.669s

With a small reordering of struct neighbour fields, subject of a
following patch, (to separate refcnt from other read mostly fields)

real 0m41.841s
user 0m15.261s
sys 8m45.949s

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

neigh: Protect neigh->ha[] with a seqlock

Add a seqlock in struct neighbour to protect neigh->ha[], and avoid
dirtying neighbour in stress situation (many different flows / dsts)

Dirtying takes place because of read_lock(&n->lock) and n->used writes.

Switching to a seqlock, and writing n->used only on jiffies changes
permits less dirtying.

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

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

Conflicts:
net/core/ethtool.c

net: clear heap allocations for privileged ethtool actions

Several other ethtool functions leave heap uncleared (potentially) by
drivers. Some interfaces appear safe (eeprom, etc), in that the sizes
are well controlled. In some situations (e.g. unchecked error conditions),
the heap will remain unchanged in areas before copying back to userspace.
Note that these are less of an issue since these all require CAP_NET_ADMIN.

Cc: stable@kernel.org
Signed-off-by: Kees Cook <kees.cook@canonical.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

NET: pch, fix use after free

Stanse found that pch_gbe_xmit_frame uses skb after it is freed. Fix
that.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Masayuki Ohtake <masa-korg@dsn.okisemi.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

NET: wimax, fix use after free

Stanse found that i2400m_rx frees skb, but still uses skb->len even
though it has skb_len defined. So use skb_len properly in the code.

And also define it unsinged int rather than size_t to solve
compilation warnings.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com>
Cc: linux-wimax@intel.com
Acked-by: Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ATM: iphase, remove sleep-inside-atomic

Stanse found that ia_init_one locks a spinlock and inside of that it
calls ia_start which calls:
* request_irq
* tx_init which does kmalloc(GFP_KERNEL)

Both of them can thus sleep and result in a deadlock. I don't see a
reason to have a per-device spinlock there which is used only there
and inited right before the lock location. So remove it completely.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

ATM: mpc, fix use after free

Stanse found that mpc_push frees skb and then it dereferences it. It
is a typo, new_skb should be dereferenced there.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ATM: solos-pci, remove use after free

Stanse found we do in console_show:
  kfree_skb(skb);
  return skb->len;
which is not good. Fix that by remembering the len and use it in the
function instead.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Chas Williams <chas@cmf.nrl.navy.mil>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

neigh: speedup neigh_hh_init()

When a new dst is used to send a frame, neigh_resolve_output() tries to
associate an struct hh_cache to this dst, calling neigh_hh_init() with
the neigh rwlock write locked.

Most of the time, hh_cache is already known and linked into neighbour,
so we find it and increment its refcount.

This patch changes the logic so that we call neigh_hh_init() with
neighbour lock read locked only, so that fast path can be run in
parallel by concurrent cpus.

This brings part of the speedup we got with commit c7d4426a98a5f
(introduce DST_NOCACHE flag) for non cached dsts, even for cached ones,
removing one of the contention point that routers hit on multiqueue
enabled machines.

Further improvements would need to use a seqlock instead of an rwlock to
protect neigh->ha[], to not dirty neigh too often and remove two atomic
ops.

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

net/fec: carrier off initially to avoid root mount failure

with hardware slow in negotiation, the system did freeze
while trying to mount root on nfs at boot time.

the link state has not been initialised so network stack
tried to start transmission right away. this caused instant
retries, as the driver solely stated business upon link down,
rendering the system unusable.

notify carrier off initially to prevent transmission until
phylib will report link up.

Signed-off-by: Oskar Schirmer <oskar@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

irda: Fix heap memory corruption in iriap.c

While parsing the GetValuebyClass command frame, we could potentially write
passed the skb->data pointer.

Cc: stable@kernel.org
Reported-by: Ilja Van Sprundel <ivansprundel@ioactive.com>
Signed-off-by: Samuel Ortiz <samuel@sortiz.org>

irda: Fix parameter extraction stack overflow

Cc: stable@kernel.org
Reported-by: Ilja Van Sprundel <ivansprundel@ioactive.com>
Signed-off-by: Samuel Ortiz <samuel@sortiz.org>

irda: Test index before read in stir421x_patch_device()

Test whether index exceeds fw->size before reading the element

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Samuel Ortiz <samuel@sortiz.org>

irda: Remove BKL instances from irnet

The code intends to lock the irnet_socket, so adding a mutex to it allows
for a complet BKL removal.

Signed-off-by: Samuel Ortiz <samuel@sortiz.org>

irda: Remove BKL instances from af_irda.c

Most of the times, lock_kernel() was pointless or could simply be replaced
by lock_sock().

Signed-off-by: Samuel Ortiz <samuel@sortiz.org>

sundance: get_stats proper locking

sundance get_stats() should not be run concurrently, add a lock to avoid
potential losses.

Note: Remove unused rx_lock field

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

net/tg3: simplify conditional

Simplify: ((a && !b) || (!a && b)) => (a != b)

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

ehea: simplify conditional

Simplify: ((a && b) || (!a && !b)) => (a == b)

Signed-off-by: Nicolas Kaiser <nikai@nikai.net>
Acked-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: use __TCA_HTB_MAX and TCA_HTB_MAX

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

r8169: use device model DMA API

Use DMA API as PCI equivalents will be deprecated. This change also
allow to allocate with GFP_KERNEL where possible.

Tested-by: Neal Becker <ndbecker2@gmail.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

r8169: allocate with GFP_KERNEL flag when able to sleep

We have fedora bug report where driver fail to initialize after
suspend/resume because of memory allocation errors:
https://bugzilla.redhat.com/show_bug.cgi?id=629158

To fix use GFP_KERNEL allocation where possible.

Tested-by: Neal Becker <ndbecker2@gmail.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Fix rxq ref counting

The rx->count reference is used to track reference counts to the
number of rx-queue kobjects created for the device.  This patch
eliminates initialization of the counter in netif_alloc_rx_queues
and instead increments the counter each time a kobject is created.
This is now symmetric with the decrement that is done when an object is
released.

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

Phonet: mark the pipe controller as EXPERIMENTAL

There are a bunch of issues that need to be fixed, including:
 - GFP_KERNEL allocations from atomic context
   (and GFP_ATOMIC in process context),
 - abuse of the setsockopt() call convention,
 - unprotected/unlocked static variables...

IMHO, we will need to alter the userspace ABI when we fix it. So mark
the configuration option as EXPERIMENTAL for the time being (or should
it be BROKEN instead?).

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: cleanup pipe enable socket option

The current code works like this:

  int garbage, status;
  socklen_t len = sizeof(status);

  /* enable pipe */
  setsockopt(fd, SOL_PNPIPE, PNPIPE_ENABLE, &garbage, sizeof(garbage));
  /* disable pipe */
  setsockopt(fd, SOL_PNPIPE, PNPIPE_DISABLE, &garbage, sizeof(garbage));
  /* get status */
  getsockopt(fd, SOL_PNPIPE, PNPIPE_INQ, &status, &len);

...which does not follow the usual socket option pattern. This patch
merges all three "options" into a single gettable&settable option,
before Linux 2.6.37 gets out:

  int status;
  socklen_t len = sizeof(status);

  /* enable pipe */
  status = 1;
  setsockopt(fd, SOL_PNPIPE, PNPIPE_ENABLE, &status, sizeof(status));
  /* disable pipe */
  status = 0;
  setsockopt(fd, SOL_PNPIPE, PNPIPE_ENABLE, &status, sizeof(status));
  /* get status */
  getsockopt(fd, SOL_PNPIPE, PNPIPE_ENABLE, &status, &len);

This also fixes the error code from EFAULT to ENOTCONN.

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

Phonet: advise against enabling the pipe controller

As it currently is, the new code path is not compatible with existing
Nokia modems. This would break existing userspace for Nokia modem, such
as the existing oFono ISI driver.

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: add to MAINTAINERS and add myself

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: update driver version 5.0.11

Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: change all P3 references to P3P

This patch just rename all P3 #define to P3P.

Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: fix promiscous mode for VF

o Allow promiscous mode setting for VF's depending upon the configuration.

Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: fix board description

Remove "Flex-10" from board description.

Signed-off-by: Sritej Velaga <sritej.velaga@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: remove private LRO flag

LRO was not getting enable after interface down/up.

Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: support quiescent mode

Put device in quiescent mode during internal loopback test.
Before running test, set state to NEED_QUISCENT. After getting
ack from all function, change state to QUISCENT and perform test.

Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qlcnic: driver private workqueue

Currently fw recovery usage global workqueue.
As same workqueue used by kernel for ethtool and etc., supporting
quiescent mode is not possible, without driver private workqueue.

Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

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

Conflicts:
Documentation/feature-removal-schedule.txt
drivers/net/wireless/ipw2x00/ipw2200.c

ipv4: Remove leftover rcu_read_unlock calls from __mkroute_output()

Commit "fib: RCU conversion of fib_lookup()" removed rcu_read_lock() from
__mkroute_output but left a couple of calls to rcu_read_unlock() in there.
This causes lockdep to complain that the rcu_read_unlock() call in
__ip_route_output_key causes a lock inbalance and quickly crashes the
kernel. The below fixes this for me.

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

net: clear heap allocation for ETHTOOL_GRXCLSRLALL

Calling ETHTOOL_GRXCLSRLALL with a large rule_cnt will allocate kernel
heap without clearing it. For the one driver (niu) that implements it,
it will leave the unused portion of heap unchanged and copy the full
contents back to userspace.

Signed-off-by: Kees Cook <kees.cook@canonical.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

sfc: Don't try to set filters with search depths we know won't work

The filter engine will time-out and ignore filters beyond
200-something hops.  We also need to avoid infinite loops in
efx_filter_search() when the table is full.

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

net: Update kernel-doc for netif_set_real_num_rx_queues()

Synchronise the comment with the preceding implementation change.

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

isdn: strcpy() => strlcpy()

setup.phone and setup.eazmsn are 32 character buffers.
rcvmsg.msg_data.byte_array is a 48 character buffer.
sc_adapter[card]->channel[rcvmsg.phy_link_no - 1].dn is 50 chars.

The rcvmsg struct comes from the memcpy_fromio() in receivemessage().
I guess that means it's data off the wire.  I'm not very familiar with
this code but I don't see any reason to assume these strings are NULL
terminated.

Also it's weird that "dn" in a 50 character buffer but we only seem to
use 32 characters.  In drivers/isdn/sc/scioc.h, "dn" is only a 49
character buffer.  So potentially there is still an issue there.

The important thing for now is to prevent the memory corruption.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ath9k_hw: Fix hw reset failure with HTC driver

The following commit removed DISABLE_REGWRITE_BUFFER ops. The unnecessary
REGWRITE_BUFFER_FLUSH was not removed properly which is causing failure on
hw reset.

Author: Felix Fietkau <nbd@openwrt.org>
Date:   Tue Oct 5 12:03:42 2010 +0200

    ath9k_hw: clean up register write buffering

Signed-off-by: Rajkumar Manoharan <rmanoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: constify WDS address

There's no need for the WDS peer address
to not be const, so make it const.

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

nl80211: use new genl helpers for WDS

Bill Jordan's patch to allow setting the WDS
peer crossed with my patch removing all the
boilerplate code in nl80211, and consequently
he didn't make use of it yet. Fix that.

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

mac80211: fix sw scan locking

The recent scan overhaul broke locking
because now we can jump to code that
attempts to unlock, while we don't have
the mutex held. Fix this by holding the
mutex around all the relevant code.

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

wl1271: Support FEM autodetection

In the NVS/INI-parameters for the driver/firmware there is the FEM
(front end module) selection option, which may configure the hardware FEM
via autodetection or manual selection. So far, there has been support for
manual selection only.

Add support for FEM autodetection.

Signed-off-by: Juuso Oikarinen <juuso.oikarinen@nokia.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

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

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

Revert "mac80211: use netif_receive_skb in ieee80211_tx_status callpath"

This reverts commit 5ed3bc7288487bd4f891f420a07319e0b538b4fe.

It turns-out that not all drivers are calling ieee80211_tx_status from a
compatible context.  Revert this for now and try again later...

Signed-off-by: John W. Linville <linville@tuxdriver.com>

Revert "rtl8180: use NAPI for bottom-half processing"

This reverts commit 030725d2c7c1fafec7ede618647bf30ed79601f0.

This commit relies on commit 5ed3bc7288487bd4f891f420a07319e0b538b4fe
("mac80211: use netif_receive_skb in ieee80211_tx_status callpath")
Unfortunately not all drivers are calling ieee80211_tx_status from a
compatible context, so that commit needs to be reverted in 2.6.36.

Signed-off-by: John W. Linville <linville@tuxdriver.com>

Merge branch 'master' of git://git./linux/kernel/git/padovan/bluetooth-2.6

ixgbe: Use affinity_hint when Flow Director is enabled

Use the new infrastructure to balance interrupts for flow
alignment when ATR or Flow Director are enabled.

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

drivers/net/stmmac/: add HAS_IOMEM dependency

The stmmac driver does not compile on s390:

drivers/net/stmmac/stmmac_main.c: In function 'stmmac_adjust_link':
drivers/net/stmmac/stmmac_main.c:210: error: implicit declaration of function 'readl'
drivers/net/stmmac/stmmac_main.c:263: error: implicit declaration of function 'writel'
drivers/net/stmmac/stmmac_main.c: In function 'stmmac_dvr_probe':
drivers/net/stmmac/stmmac_main.c:1674: error: implicit declaration of function 'ioremap'
drivers/net/stmmac/stmmac_main.c:1674: warning: assignment makes pointer from integer without a cast
drivers/net/stmmac/stmmac_main.c:1761: error: implicit declaration of function 'iounmap'
make[3]: *** [drivers/net/stmmac/stmmac_main.o] Error 1

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Giuseppe CAVALLARO <peppe.cavallaro@st.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: netif_set_real_num_rx_queues may cap num_rx_queues at init time

Do not set num_rx_queues in netif_set_real_num_rx_queues() some
drivers will increase the real_num_rx_queues later due to a feature
changes or available interrupts increasing. By setting num_rx_queues
here this ends up creating a cap on the number of rx queues
available.

For example the ixgbe driver sets the max number of queues it intends
to use ever then sets the current number in use with the
netif_set_num_{rx|tx}_queues calls. With the current implementation
the number of rx queues gets limited so when a feature such as DCB
or FCoE is enabled the queues are no longer available.

kobjects will only be allocated for real_num_rx_queues so the waste
in memory is minimal.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

eicon: make buffer larger

In diva_mnt_add_xdi_adapter() we do this:
  strcpy (clients[id].drvName,     tmp);
  strcpy (clients[id].Dbg.drvName, tmp);

The "clients[id].drvName" is a 128 character buffer and
"clients[id].Dbg.drvName" was originally a 16 character buffer but I've
changed it to 128 as well.  We don't actually use 128 characters but we
do use more than 16.

I've also changed the size of "tmp" to 128 characters instead of 256.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

dccp: Kill dead code and add static markers.

Remove dead code and make some functions static.
Compile tested only.

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

sundance: Drop legacy _COMPAT_WITH_OLD_KERNEL includes

Drop legacy includes since 2.4.x

Signed-off-by: Denis Kirjanov <dkirjanov@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

TCP: Fix setting of snd_ssthresh in tcp_mtu_probe_success

This looks like a simple typo that has gone unnoticed for some time.  The
impact is relatively low but it's clearly wrong.

Signed-off-by: John Heffner <johnwheffner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

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

Conflicts:
drivers/net/pcmcia/pcnet_cs.c
net/caif/caif_socket.c

Merge branch 'master' of /linux/kernel/git/torvalds/linux-2.6

bonding: add Speed/Duplex information to /proc/net/bonding/bond

Effect:
 Slave Interface: eth5
 MII Status: up
 Speed: 10000 Mbps
 Duplex: full
 Link Failure Count: 0
 Permanent HW addr: XX:XX:XX:XX:XX:XX
 Slave queue ID: 0

Signed-off-by: Krzysztof Piotr Oledzki <ole@ans.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

neigh: RCU conversion of struct neighbour

This is the second step for neighbour RCU conversion.

(first was commit d6bf7817 : RCU conversion of neigh hash table)

neigh_lookup() becomes lockless, but still take a reference on found
neighbour. (no more read_lock()/read_unlock() on tbl->lock)

struct neighbour gets an additional rcu_head field and is freed after an
RCU grace period.

Future work would need to eventually not take a reference on neighbour
for temporary dst (DST_NOCACHE), but this would need dst->_neighbour to
use a noref bit like we did for skb->_dst.

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

iwlagn: add define for MODULE_FIRMWARE

Adding MODULE_FIRMWARE define for 130 series of devices

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>

iwlagn: fix module firmware name for 130 series device

130 series device should use firmware name with 130 in it.

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>

bonding: reread information about speed and duplex when interface goes up

When an interface was enslaved when it was down, bonding thinks
it has speed -1 even after it goes up. This leads into selecting
a wrong active interface in active/backup mode on mixed 10G/1G or
1G/100M environment.

before:
 bonding: bond0: link status definitely up for interface eth5, 100 Mbps full duplex.
 bonding: bond0: link status definitely up for interface eth0, 100 Mbps full duplex.

after:
 bonding: bond0: link status definitely up for interface eth5, 10000 Mbps full duplex.
 bonding: bond0: link status definitely up for interface eth0, 1000 Mbps full duplex.

Signed-off-by: Krzysztof Piotr Oledzki <ole@ans.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

bonding: print information about speed and duplex seen by the driver

before:
 bonding: bond0: link status definitely up for interface eth5
 bonding: bond0: link status definitely up for interface eth0

after:
 bonding: bond0: link status definitely up for interface eth5, 100 Mbps full duplex.
 bonding: bond0: link status definitely up for interface eth0, 100 Mbps full duplex.

Signed-off-by: Krzysztof Piotr Oledzki <ole@ans.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: update version to 1.60.00-1

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

bnx2x: properly initialize FW stats

Client statistics need to be initialized to -1

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

bnx2x: code beautify

This patch does not include any functional changes.
The changes are: empty lines, indentation and comments.

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

bnx2x, cnic: Fix SPQ return credit

Return proper L2 and L5 SPQ (slow path queue) credits.  Previously, all
SPQ events were counted as L5 types.

Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: move msix table initialization to probe()

Decide which interrupt mode to use (MSI-X, MSI, INTa) only once in probe() and
initialize appropriate structures.

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

bnx2x: use L1_CACHE_BYTES instead of magic number

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

bnx2x: use proper constants for dma_unmap* calls

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

bnx2x: remove unused fields in main driver structure

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

bnx2x: remove unused parameter in reuse_rx_skb()

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

bnx2x: Add 57712 support

57712 HW supported with same set of features as for 57710/57711

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

bnx2x: change type of spq_left to atomic

The field is now accessed from different contexts.

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

bnx2x: rename MF related fields

MF (multi-function) mode supported not only by 57711E (E1H) devices,
but also by coming 57712E, then we use more generic names.

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

bnx2x: remove old FW files

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

bnx2x, cnic, bnx2i: use new FW/HSI

This is the new FW HSI blob and the relevant definitions without logic changes.
It also included code adaptation for new HSI. New features are not enabled.

New FW/HSI includes:
- Support for 57712 HW
- Future support for VF (not used)
- Improvements in FW interrupts scheme
- FW FCoE hooks (stubs for future usage)

Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: add 6.0.34 fw files

New firmware files for all bnx2x devices

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

bnx2x: create folder for bnx2x firmware files

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

cnic: Fine-tune ring init code.

Initialize the rings only during cnic_uio_open() and shut them down
during cnic_uio_close().  Check for the new bit CNIC_LCL_FL_RINGS_INITED
before checking for ring interrupt.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cnic: Use pfid for internal memory offsets.

The pfid (physical function ID) is the same as PCI function on production
devices.  The pfid for future devices will be different and will be used
for internal memory offsets.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cnic: Pass cp pointer to BNX2X_HW_CID.

Preparation work for upcoming firmware interface changes.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PATCH net-next] tg3: Set real_num_rx_queues for non-multiq devs

Commit 2ddaad397c47de012dfb956b0c05540da1a0dde5 entitled "tg3: Use
netif_set_real_num_{rx,tx}_queues()" added a new call to
netif_set_real_num_rx_queues in tg3_enable_msix().  This call also needs
to be added to the legacy path to correctly reflect the actual number of
rx queues.

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

mac80211: avoid uninitialized var warning in ieee80211_scan_cancel

net/mac80211/scan.c: In function ‘ieee80211_scan_cancel’:
net/mac80211/scan.c:794: warning: ‘finish’ may be used uninitialized in this function

Signed-off-by: John W. Linville <linville@tuxdriver.com>