This reverts commits 9a036b93 ("x86/signal/64: Remove 'fs' and 'gs'
from sigcontext") and c6f20629 ("x86/signal/64: Fix SS handling for
signals delivered to 64-bit programs").

They were cleanups, but they break dosemu by changing the signal return
behavior (and removing 'fs' and 'gs' from the sigcontext struct - while
not actually changing any behavior - causes build problems).
Reported-and-tested-by: Stas Sergeev <stsp@list.ru>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull networking fixes from David Miller:

 1) Workaround hw bug when acquiring PCI bos ownership of iwlwifi
    devices, from Emmanuel Grumbach.

 2) Falling back to vmalloc in conntrack should not emit a warning, from
    Pablo Neira Ayuso.

 3) Fix NULL deref when rtlwifi driver is used as an AP, from Luis
    Felipe Dominguez Vega.

 4) Rocker doesn't free netdev on device removal, from Ido Schimmel.

 5) UDP multicast early sock demux has route handling races, from Eric
    Dumazet.

 6) Fix L4 checksum handling in openvswitch, from Glenn Griffin.

 7) Fix use-after-free in skb_set_peeked, from Herbert Xu.

 8) Don't advertize NETIF_F_FRAGLIST in virtio_net driver, this can lead
    to fraglists longer than the driver can support.  From Jason Wang.

 9) Fix mlx5 on non-4k-pagesize systems, from Carol L Soto.

10) Fix interrupt storm in bna driver, from Ivan Vecera.

11) Don't propagate -EBUSY from netlink_insert(), from Daniel Borkmann.

12) Fix inet request sock leak, from Eric Dumazet.

13) Fix TX interrupt masking and marking in TX descriptors of fs_enet
    driver, from LEROY Christophe.

14) Get rid of rule optimizer in gianfar driver, it's buggy and unlikely
    to get fixed any time soon.  From Jakub Kicinski

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (61 commits)
  cosa: missing error code on failure in probe()
  gianfar: remove faulty filer optimizer
  gianfar: correct list membership accounting
  gianfar: correct filer table writing
  bonding: Gratuitous ARP gets dropped when first slave added
  net: dsa: Do not override PHY interface if already configured
  net: fs_enet: mask interrupts for TX partial frames.
  net: fs_enet: explicitly remove I flag on TX partial frames
  inet: fix possible request socket leak
  inet: fix races with reqsk timers
  mkiss: Fix error handling in mkiss_open()
  bnx2x: Free NVRAM lock at end of each page
  bnx2x: Prevent null pointer dereference on SKB release
  cxgb4: missing curly braces in t4_setup_debugfs()
  net-timestamp: Update skb_complete_tx_timestamp comment
  ipv6: don't reject link-local nexthop on other interface
  netlink: make sure -EBUSY won't escape from netlink_insert
  bna: fix interrupts storm caused by erroneous packets
  net: mvpp2: replace TX coalescing interrupts with hrtimer
  net: mvpp2: enable proper per-CPU TX buffers unmapping
  ...

Pull EDAC fix from Borislav Petkov:
 "A ppc4xx_edac fix for accessing ->csrows properly.  This driver was
  missed during the conversion a couple of years ago"

* tag 'edac_fix_for_4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp:
  EDAC, ppc4xx: Access mci->csrows array elements properly

The commit

  de3910eb ("edac: change the mem allocation scheme to
		 make Documentation/kobject.txt happy")

changed the memory allocation for the csrows member. But ppc4xx_edac was
forgotten in the patch. Fix it.
Signed-off-by: Michael Walle <michael@walle.cc>
Cc: <stable@vger.kernel.org>
Cc: linux-edac <linux-edac@vger.kernel.org>
Cc: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Link: http://lkml.kernel.org/r/1437469253-8611-1-git-send-email-michael@walle.ccSigned-off-by: Borislav Petkov <bp@suse.de>

If register_hdlc_device() fails, the current code returns 0 but we
should return an error code instead.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jakub Kicinski says:

====================
gianfar: filer changes

respinning with examples as requested.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Current filer rule optimization is broken in several ways:
 (1) Can perform reads/writes beyond end of allocated tables.
     (gianfar_ethtool.c:1326).

(2) It breaks badly for rules with more than 2 specifiers
     (e.g. matching ip, port, tos).

Example:
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.1 dst-port 1 tos 1 action 1
Added rule with ID 254
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.2 dst-port 2 tos 2 action 9
Added rule with ID 253
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.3 dst-port 3 tos 3 action 17
Added rule with ID 252
# ./filer_decode /sys/kernel/debug/gfar1/filer_raw
00: MASK == 00000210 AND         Q:00           ctrl:00000080 prop:00000210
01: FPR  == 00000210 AND CLE     Q:00           ctrl:00000281 prop:00000210
02: MASK == ffffffff AND         Q:00           ctrl:00000080 prop:ffffffff
03: DPT  == 00000003 AND         Q:00           ctrl:0000008e prop:00000003
04: TOS  == 00000003 AND         Q:00           ctrl:0000008a prop:00000003
05: DIA  == 0a000003 AND         Q:11           ctrl:0000448c prop:0a000003
06: DPT  == 00000002 AND         Q:00           ctrl:0000008e prop:00000002
07: TOS  == 00000002 AND         Q:00           ctrl:0000008a prop:00000002
08: DIA  == 0a000002 AND         Q:09           ctrl:0000248c prop:0a000002
09: DIA  == 0a000001 AND         Q:00           ctrl:0000008c prop:0a000001
0a: DPT  == 00000001 AND         Q:00           ctrl:0000008e prop:00000001
0b: TOS  == 00000001     CLE     Q:01           ctrl:0000060a prop:00000001
ff: MASK >= 00000000             Q:00           ctrl:00000020 prop:00000000

(Entire cluster gets AND-ed together).

 (3) We observed that the masking rules it generates do not
     play well with clustering on P2020.  Only first rule
     of the cluster would ever fire.  Given that optimizer
     relies heavily on masking this is very hard to fix.

Example:
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.1 dst-port 1  action 1
Added rule with ID 254
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.2 dst-port 2  action 9
Added rule with ID 253
# ethtool -N eth2 flow-type udp4 dst-ip 10.0.0.3 dst-port 3  action 17
Added rule with ID 252
# ./filer_decode /sys/kernel/debug/gfar1/filer_raw
00: MASK == 00000210 AND         Q:00           ctrl:00000080 prop:00000210
01: FPR  == 00000210 AND CLE     Q:00           ctrl:00000281 prop:00000210
02: MASK == ffffffff AND         Q:00           ctrl:00000080 prop:ffffffff
03: DPT  == 00000003 AND         Q:00           ctrl:0000008e prop:00000003
04: DIA  == 0a000003             Q:11           ctrl:0000440c prop:0a000003
05: DPT  == 00000002 AND         Q:00           ctrl:0000008e prop:00000002
06: DIA  == 0a000002             Q:09           ctrl:0000240c prop:0a000002
07: DIA  == 0a000001 AND         Q:00           ctrl:0000008c prop:0a000001
08: DPT  == 00000001     CLE     Q:01           ctrl:0000060e prop:00000001
ff: MASK >= 00000000             Q:00           ctrl:00000020 prop:00000000

Which looks correct according to the spec but only the first
(eth id 252)/last added rule for 10.0.0.3 will ever trigger.
As if filer did not treat the AND CLE as cluster start but
also kept AND-ing the rules.  We found no errata covering this.

The fact that nobody noticed (2) or (3) makes me think
that this feature is not very widely used and we should just
remove it.
Reported-by: Aleksander Dutkowski <adutkowski@gmail.com>
Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Acked-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

At a cost of one line let's make sure .count is correct
when calling gfar_process_filer_changes().
Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

MAX_FILER_IDX is the last usable index.  Using less-than
will already guarantee that one entry for catch-all rule
will be left, no need to subtract 1 here.
Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>

When the first slave is added (such as during bootup) the first
gratuitous ARP gets dropped. We don't see this drop during a failover.
The packet gets dropped in qdisc (noop_enqueue).

The fix is to delay the sending of gratuitous ARPs till the bond dev's
carrier is present.

It can also be worked around by setting num_grat_arp to more than 1.
Signed-off-by: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In case we need to divert reads/writes using the slave MII bus, we may have
already fetched a valid PHY interface property from Device Tree, and that
mode is used by the PHY driver to make configuration decisions.

If we could not fetch the "phy-mode" property, we will assign p->phy_interface
to PHY_INTERFACE_MODE_NA, such that we can actually check for that condition as
to whether or not we should override the interface value.

Fixes: 19334920 ("net: dsa: Set valid phy interface type")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull arm64 fix from Catalin Marinas:
 "Fix coarse clock monotonicity (VDSO timestamp off by one jiffy
  compared to the syscall one)"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: VDSO: fix coarse clock monotonicity regression

Pull amd drm fixes from Alex Deucher:
 "Dave is on vacation at the moment, so please pull these radeon and
  amdgpu fixes directly.

  Just a few minor things for 4.2:

   - add a new radeon pci id
   - fix a power management regression in amdgpu
   - fix HEVC command buffer validation in amdgpu"

* 'drm-fixes-4.2' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: add new OLAND pci id
  Revert "drm/amdgpu: Configure doorbell to maximum slots"
  drm/amdgpu: add context buffer size check for HEVC

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

This reverts commit 78ad5cdd.
This commit breaks dpm and suspend/resume on CZ.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

Pull regmap fix from Mark Brown:
 "regmap: Fix handling of present bits on rbtree cache block resize

  When expanding a cache block we use krealloc() to resize the register
  present bitmap without initialising the newly allocated data (the
  original code was written for kzalloc()).  Add an appropraite memset()
  to fix that"

* tag 'regmap-fix-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: regcache-rbtree: Clean new present bits on present bitmap resize

Merge tag 'localmodconfig-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-kconfig

Pull localmodconfig fix from Steven Rostedt:
 "Leonidas Spyropoulos found that modules like nouveau were being
  unselected by make localmodconfig even though their configs were set
  and the module was loaded and visible by lsmod.

  The reason for this was because streamline-config.pl only looks at
  Makefiles, and not Kbuild files.  As these modules use Kbuild for
  their names, they too need to be checked by localmodconfig.  This was
  fixed by Richard Weinberger"

* tag 'localmodconfig-v4.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-kconfig:
  localmodconfig: Use Kbuild files too

In kbuild it is allowed to define objects in files named "Makefile"
and "Kbuild".
Currently localmodconfig reads objects only from "Makefile"s and misses
modules like nouveau.

Link: http://lkml.kernel.org/r/1437948415-16290-1-git-send-email-richard@nod.at

Cc: stable@vger.kernel.org
Reported-and-tested-by: Leonidas Spyropoulos <artafinde@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

Johan Hedberg says:

====================
pull request: bluetooth 2015-08-11

Here's an important regression fix for the 4.2-rc series that ensures
user space isn't given invalid LTK values. The bug essentially prevents
the encryption of subsequent LE connections, i.e. makes it impossible to
pair devices over LE.

Let me know if there are any issues pulling. Thanks.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

We are not interested in interrupts for partially transmitted frames.
Unlike SCC and FCC, the FEC doesn't handle the I bit in buffer
descriptors, instead it defines two interrupt bits, TXB and TXF.

We have to mask TXB in order to only get interrupts once the
frame is fully transmitted.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

We are not interested in interrupts for partially transmitted frames,
we have to clear BD_ENET_TX_INTR explicitly otherwise it may remain
from a previously used descriptor.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull fbdev fixes from Tomi Valkeinen:
 - fix display regression on Versatile boards
 - fix OF node refcount bugs on omapdss
 - fix WARN about clock prepare on pxa3xx_gcu
 - fix mem leak in videomode helpers
 - fix fbconsole related boot problem on sun7i-a20-olinuxino-micro

* tag 'fbdev-fixes-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux:
  fbcon: unconditionally initialize cursor blink interval
  video: Fix possible leak in of_get_videomode()
  video: fbdev: pxa3xx_gcu: prepare the clocks
  OMAPDSS: Fix omap_dss_find_output_by_port_node() port refcount decrement
  OMAPDSS: Fix node refcount leak in omapdss_of_get_next_port()
  fbdev: select versatile helpers for the integrator

In commit b357a364 ("inet: fix possible panic in
reqsk_queue_unlink()"), I missed fact that tcp_check_req()
can return the listener socket in one case, and that we must
release the request socket refcount or we leak it.

Tested:

 Following packetdrill test template shows the issue

0     socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0    setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0    bind(3, ..., ...) = 0
+0    listen(3, 1) = 0

+0    < S 0:0(0) win 2920 <mss 1460,sackOK,nop,nop>
+0    > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK>
+.002 < . 1:1(0) ack 21 win 2920
+0    > R 21:21(0)

Fixes: b357a364 ("inet: fix possible panic in reqsk_queue_unlink()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

reqsk_queue_destroy() and reqsk_queue_unlink() should use
del_timer_sync() instead of del_timer() before calling reqsk_put(),
otherwise we could free a req still used by another cpu.

But before doing so, reqsk_queue_destroy() must release syn_wait_lock
spinlock or risk a dead lock, as reqsk_timer_handler() might
need to take this same spinlock from reqsk_queue_unlink() (called from
inet_csk_reqsk_queue_drop())

Fixes: fa76ce73 ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If register_netdev() fails we are not propagating the error and
we return success because ax_open() succeeded previously.

Fix this by checking the return value of ax_open() and
register_netdev() and propagate the error in case of failure.
Reported-by: RUC_Soft_Sec <zy900702@163.com>
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

The following patchset contains five Netfilter fixes for your net tree,
they are:

1) Silence a warning on falling back to vmalloc(). Since 88eab472, we can
   easily hit this warning message, that gets users confused. So let's get rid
   of it.

2) Recently when porting the template object allocation on top of kmalloc to
   fix the netns dependencies between x_tables and conntrack, the error
   checks where left unchanged. Remove IS_ERR() and check for NULL instead.
   Patch from Dan Carpenter.

3) Don't ignore gfp_flags in the new nf_ct_tmpl_alloc() function, from
   Joe Stringer.

4) Fix a crash due to NULL pointer dereference in ip6t_SYNPROXY, patch from
   Phil Sutter.

5) The sequence number of the Syn+ack that is sent from SYNPROXY to clients is
   not adjusted through our NAT infrastructure, as a result the client may
   ignore this TCP packet and TCP flow hangs until the client probes us.  Also
   from Phil Sutter.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull HID fixes from Jiri Kosina:

 - fix for bounds limit calculation in uclogic driver, by Dan Carpenter

 - fix for use-after-free during device removal, by Krzysztof Kozlowski

 - fix for userspace regression (that became apparent only with shiny
   new libinput, so it's not that bad, but I still consider it 4.2
   material), in wacom driver, by Jason Gerecke

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid:
  HID: wacom: Report correct device resolution when using the wireless adapater
  HID: hid-input: Fix accessing freed memory during device disconnect
  HID: uclogic: fix limit in uclogic_tablet_enable()

The 'wacom_wireless_work' function does not recalculate the tablet's
resolution, causing the value contained in the 'features' struct to
always be reported to userspace. This value is valid only for the pen
interface, meaning that the value will be incorrect for the touchpad (if
present). This in particular causes problems for libinput which relies
on the reported resolution being correct.

This patch adds the necessary calls to recalculate the resolution for
each interface. This requires a little bit of code shuffling since both
the 'wacom_set_default_phy' and 'wacom_calculate_res' are declared below
their new first point of use in 'wacom_wireless_work'.
Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>

Yuval Mintz says:

====================
bnx2x: small fixes

This adds 2 small fixes, one to error flows during memory release
and the other to flash writes via ethtool API.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Writing each 4Kb page into flash might take up-to ~100 miliseconds,
during which time management firmware cannot acces the nvram for its
own uses.

Firmware upgrade utility use the ethtool API to burn new flash images
for the device via the ethtool API, doing so by writing several page-worth
of data on each command. Such action might create problems for the
management firmware, as the nvram might not be accessible for a long time.

This patch changes the write implementation, releasing the nvram lock on
the completion of each page, allowing the management firmware time to
claim it and perform its own required actions.
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Ariel Elior <Ariel.Elior@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On error flows its possible to free an SKB even if it was not allocated.
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Ariel Elior <Ariel.Elior@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There were missing curly braces so it means we call add_debugfs_mem()
unintentionally.

Fixes: 3ccc6cf7 ('cxgb4: Adds support for T6 adapter')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

After "62bccb8c net-timestamp: Make the clone operation stand-alone from phy
timestamping" the hwtstamps parameter of skb_complete_tx_timestamp() may no
longer be NULL.
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Cc: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

48ed7b26 ("ipv6: reject locally assigned nexthop addresses") is too
strict; it rejects following corner-case:

ip -6 route add default via fe80::1:2:3 dev eth1

[ where fe80::1:2:3 is assigned to a local interface, but not eth1 ]

Fix this by restricting search to given device if nh is linklocal.

Joint work with Hannes Frederic Sowa.

Fixes: 48ed7b26 ("ipv6: reject locally assigned nexthop addresses")
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Linus reports the following deadlock on rtnl_mutex; triggered only
once so far (extract):

[12236.694209] NetworkManager  D 0000000000013b80     0  1047      1 0x00000000
[12236.694218]  ffff88003f902640 0000000000000000 ffffffff815d15a9 0000000000000018
[12236.694224]  ffff880119538000 ffff88003f902640 ffffffff81a8ff84 00000000ffffffff
[12236.694230]  ffffffff81a8ff88 ffff880119c47f00 ffffffff815d133a ffffffff81a8ff80
[12236.694235] Call Trace:
[12236.694250]  [<ffffffff815d15a9>] ? schedule_preempt_disabled+0x9/0x10
[12236.694257]  [<ffffffff815d133a>] ? schedule+0x2a/0x70
[12236.694263]  [<ffffffff815d15a9>] ? schedule_preempt_disabled+0x9/0x10
[12236.694271]  [<ffffffff815d2c3f>] ? __mutex_lock_slowpath+0x7f/0xf0
[12236.694280]  [<ffffffff815d2cc6>] ? mutex_lock+0x16/0x30
[12236.694291]  [<ffffffff814f1f90>] ? rtnetlink_rcv+0x10/0x30
[12236.694299]  [<ffffffff8150ce3b>] ? netlink_unicast+0xfb/0x180
[12236.694309]  [<ffffffff814f5ad3>] ? rtnl_getlink+0x113/0x190
[12236.694319]  [<ffffffff814f202a>] ? rtnetlink_rcv_msg+0x7a/0x210
[12236.694331]  [<ffffffff8124565c>] ? sock_has_perm+0x5c/0x70
[12236.694339]  [<ffffffff814f1fb0>] ? rtnetlink_rcv+0x30/0x30
[12236.694346]  [<ffffffff8150d62c>] ? netlink_rcv_skb+0x9c/0xc0
[12236.694354]  [<ffffffff814f1f9f>] ? rtnetlink_rcv+0x1f/0x30
[12236.694360]  [<ffffffff8150ce3b>] ? netlink_unicast+0xfb/0x180
[12236.694367]  [<ffffffff8150d344>] ? netlink_sendmsg+0x484/0x5d0
[12236.694376]  [<ffffffff810a236f>] ? __wake_up+0x2f/0x50
[12236.694387]  [<ffffffff814cad23>] ? sock_sendmsg+0x33/0x40
[12236.694396]  [<ffffffff814cb05e>] ? ___sys_sendmsg+0x22e/0x240
[12236.694405]  [<ffffffff814cab75>] ? ___sys_recvmsg+0x135/0x1a0
[12236.694415]  [<ffffffff811a9d12>] ? eventfd_write+0x82/0x210
[12236.694423]  [<ffffffff811a0f9e>] ? fsnotify+0x32e/0x4c0
[12236.694429]  [<ffffffff8108cb70>] ? wake_up_q+0x60/0x60
[12236.694434]  [<ffffffff814cba09>] ? __sys_sendmsg+0x39/0x70
[12236.694440]  [<ffffffff815d4797>] ? entry_SYSCALL_64_fastpath+0x12/0x6a

It seems so far plausible that the recursive call into rtnetlink_rcv()
looks suspicious. One way, where this could trigger is that the senders
NETLINK_CB(skb).portid was wrongly 0 (which is rtnetlink socket), so
the rtnl_getlink() request's answer would be sent to the kernel instead
to the actual user process, thus grabbing rtnl_mutex() twice.

One theory would be that netlink_autobind() triggered via netlink_sendmsg()
internally overwrites the -EBUSY error to 0, but where it is wrongly
originating from __netlink_insert() instead. That would reset the
socket's portid to 0, which is then filled into NETLINK_CB(skb).portid
later on. As commit d470e3b4 ("[NETLINK]: Fix two socket hashing bugs.")
also puts it, -EBUSY should not be propagated from netlink_insert().

It looks like it's very unlikely to reproduce. We need to trigger the
rhashtable_insert_rehash() handler under a situation where rehashing
currently occurs (one /rare/ way would be to hit ht->elasticity limits
while not filled enough to expand the hashtable, but that would rather
require a specifically crafted bind() sequence with knowledge about
destination slots, seems unlikely). It probably makes sense to guard
__netlink_insert() in any case and remap that error. It was suggested
that EOVERFLOW might be better than an already overloaded ENOMEM.

Reference: http://thread.gmane.org/gmane.linux.network/372676Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

The commit "e29aa339 bna: Enable Multi Buffer RX" moved packets counter
increment from the beginning of the NAPI processing loop after the check
for erroneous packets so they are never accounted. This counter is used
to inform firmware about number of processed completions (packets).
As these packets are never acked the firmware fires IRQs for them again
and again.

Fixes: e29aa339 ("bna: Enable Multi Buffer RX")
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Rasesh Mody <rasesh.mody@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Marcin Wojtas says:

====================
Fixes for the network driver of Marvell Armada 375 SoC

This is a set of three patches that fix long-lasting problems implemented in
the initial support for the Armada 375 network controller.

Due to an inappropriate concept of handling the per-CPU sent packets'
processing on TX path the driver numerous problems occured, such as RCU
stalls. Those have been fixed, of which details you can find in the commit
logs. The patches were intensively tested on top of v4.2-rc5.

I'm looking forward to any comments or remarks.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The PP2 controller is capable of per-CPU TX processing, which means there are
per-CPU banked register sets and queues. Current version of the driver supports
TX packet coalescing - once on given CPU sent packets amount reaches a threshold
value, an IRQ occurs. However, there is a single interrupt line responsible for
CPU0/1 TX and RX events (the latter is not per-CPU, the hardware does not
support RSS).

When the top-half executes the interrupt cause is not known. This is why in
NAPI poll function, along with RX processing, IRQ cause register on both
CPU's is accessed in order to determine on which of them the TX coalescing
threshold might have been reached. Thus the egress processing and releasing the
buffers is able to take place on the corresponding CPU. Hitherto approach lead
to an illegal usage of on_each_cpu function in softirq context.

The problem is solved by resigning from TX coalescing interrupts and separating
egress finalization from NAPI processing. For that purpose a method of using
hrtimer is introduced. In main transmit function (mvpp2_tx) buffers are released
once a software coalescing threshold is reached. In case not all the data is
processed a timer is set on this CPU - in its interrupt context a tasklet is
scheduled in which all queues are processed. At once only one timer per-CPU can
be running, which is controlled by a dedicated flag.

This commit removes TX processing from NAPI polling function, disables hardware
coalescing and enables hrtimer with tasklet, using new per-CPU port structure
(mvpp2_port_pcpu).
Signed-off-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mvpp2 driver allows usage of per-CPU TX processing. Once the packets are
prepared independetly on each CPU, the hardware enqueues the descriptors in
common TX queue. After they are sent, the buffers and associated sk_buffs
should be released on the corresponding CPU.

This is why a special index is maintained in order to point to the right data to
be released after transmission takes place. Each per-CPU TX queue comprise an
array of sent sk_buffs, freed in mvpp2_txq_bufs_free function. However, the
index was used there also for obtaining a descriptor (and therefore a buffer to
be DMA-unmapped) from common TX queue, which was wrong, because it was not
referring to the current CPU.

This commit enables proper unmapping of sent data buffers by indexing them in
per-CPU queues using a dedicated array for keeping their physical addresses.
Signed-off-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>