vxlan: Update tx_errors statistics if vxlan_build_skb return err.

If vxlan_build_skb return err < 0, tx_errors should be also increased.

Signed-off-by: Haishuang Yan <yanhaishuang@cmss.chinamobile.com>
Acked-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/mlx4_en: protect ring->xdp_prog with rcu_read_lock

Depending on the preempt mode, the bpf_prog stored in xdp_prog may be
freed despite the use of call_rcu inside bpf_prog_put. The situation is
possible when running in PREEMPT_RCU=y mode, for instance, since the rcu
callback for destroying the bpf prog can run even during the bh handling
in the mlx4 rx path.

Several options were considered before this patch was settled on:

Add a napi_synchronize loop in mlx4_xdp_set, which would occur after all
of the rings are updated with the new program.
This approach has the disadvantage that as the number of rings
increases, the speed of update will slow down significantly due to
napi_synchronize's msleep(1).

Add a new rcu_head in bpf_prog_aux, to be used by a new bpf_prog_put_bh.
The action of the bpf_prog_put_bh would be to then call bpf_prog_put
later. Those drivers that consume a bpf prog in a bh context (like mlx4)
would then use the bpf_prog_put_bh instead when the ring is up. This has
the problem of complexity, in maintaining proper refcnts and rcu lists,
and would likely be harder to review. In addition, this approach to
freeing must be exclusive with other frees of the bpf prog, for instance
a _bh prog must not be referenced from a prog array that is consumed by
a non-_bh prog.

The placement of rcu_read_lock in this patch is functionally the same as
putting an rcu_read_lock in napi_poll. Actually doing so could be a
potentially controversial change, but would bring the implementation in
line with sk_busy_loop (though of course the nature of those two paths
is substantially different), and would also avoid future copy/paste
problems with future supporters of XDP. Still, this patch does not take
that opinionated option.

Testing was done with kernels in either PREEMPT_RCU=y or
CONFIG_PREEMPT_VOLUNTARY=y+PREEMPT_RCU=n modes, with neither exhibiting
any drawback. With PREEMPT_RCU=n, the extra call to rcu_read_lock did
not show up in the perf report whatsoever, and with PREEMPT_RCU=y the
overhead of rcu_read_lock (according to perf) was the same before/after.
In the rx path, rcu_read_lock is eventually called for every packet
from netif_receive_skb_internal, so the napi poll call's rcu_read_lock
is easily amortized.

v2:
Remove extra rcu_read_lock in mlx4_en_process_rx_cq body
Annotate xdp_prog with __rcu, and convert all usages to rcu_assign or
rcu_dereference[_protected] as appropriate.
Add explicit mutex lock around rcu_assign instead of xchg loop.

Fixes: d576acf0a22 ("net/mlx4_en: add page recycle to prepare rx ring for tx support")
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Brenden Blanco <bblanco@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'mediatek-rx-path-enhancements'

Sean Wang says:

====================
net: ethernet: mediatek: add enhancements to RX path

Changes since v1:
- fix message typos and add coverletter

Changes since v2:
- split from the previous series for submitting add enhancements as
a series targeting 'net-next' and add indents before comments.

Changes since v3:
- merge the patch using PDMA RX path
- fixed the input of mtk_poll_rx is with the remaining budget

Changes since v4:
- save one wmb and register update when no packet is being handled
inside mtk_poll_rx call
- fixed incorrect return packet count from mtk_napi_rx
====================

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

net: ethernet: mediatek: enhance RX path by aggregating more SKBs into NAPI

The patch adds support for aggregating more SKBs feed into NAPI in
order to get more benefits from generic receive offload (GRO) by
peeking at the RX ring status and moving more packets right before
returning from NAPI RX polling handler if NAPI budgets are still
available and some packets already present in hardware.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ethernet: mediatek: enhance RX path by reducing the frequency of the memory barrier used

The patch makes move wmb() to outside the loop that could help
RX path handling more faster although that RX descriptors aren't
freed for DMA to use as soon as possible, but based on my experiment
and the result shows it still can reach about 943Mbpis without
performance drop that is tested based on the setup with one port
using Giga PHY and 256 RX descriptors for DMA to move.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'hso-neatening'

Joe Perches says:

====================
hso: neatening

This seems to be the only code in the kernel that uses
macro defines with a trailing underscore.  Fix that.
====================

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

hso: Convert printk to pr_<level>

Use a more common logging style

Miscellanea:

o Add pr_fmt to prefix each output message
o Realign arguments

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

hso: Use a more common logging style

Macros that end in an underscore are just odd.
Add hso_dbg(level, fmt, ...) and use it everwhere instead.

Several uses had additional unnecessary newlines as the
macro added a newline.  Remove the newline from the macro
and add newlines to each use as appropriate.

Remove now unused D<digit> macros.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

smsc95xx: Add mdix control via ethtool

Add mdix control through ethtool.

Signed-off-by: Woojung Huh <Woojung.huh@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

smsc95xx: Add register define

Add STRAP_STATUS defines.

Signed-off-by: Woojung Huh <Woojung.huh@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

smsc95xx: Add maintainer

Add Microchip Linux Driver Support as maintainer
because this driver is maintaining by Microchip.

Signed-off-by: Woojung Huh <Woojung.huh@gmail.com>
Acked-by: Steve Glendinning <steve.glendinning@shawell.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'mv88e6xxx-isolate-Global2'

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: isolate Global2 support

Registers of Marvell chips are organized in internal SMI devices.

One of them at address 0x1C is called Global2. It provides an extended
set of registers, used for interrupt control, EEPROM access, indirect
PHY access (to bypass the PHY Polling Unit) and cross-chip setup.

Most chips have it, but some others don't (older ones such as 6060).

Now that its related code is isolated in mv88e6xxx_g2_* functions, move
it to its own global2.c file, making most of its setup code static.

Then make its compilation optional, which allows to reduce the size of
the mv88e6xxx driver for devices such as home routers embedding Ethernet
chips without Global2 support.

It is present on most recent chips, thus enable its support by default.

Changes in v2: fail probe if GLOBAL2 is required but not enabled.
====================

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

net: dsa: mv88e6xxx: make global2 code optional

Since not every chip has a Global2 set of registers, make its support
optional, in which case the related functions will return -EOPNOTSUPP.

This also allows to reduce the size of the mv88e6xxx driver for devices
such as home routers embedding Ethernet chips without Global2 support.

It is present on most recent chips, thus enable its support by default.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: move Global2 code

Marvell chips are composed of multiple SMI devices. One of them at
address 0x1C is called Global2. It provides an extended set of
registers, used for interrupt control, EEPROM access, indirect PHY
access (to bypass the PHY Polling Unit) and cross-chip related setup.

Most chips have it, but some others don't (older ones such as 6060).

Now that its related code is isolated in mv88e6xxx_g2_* functions, move
it to its own global2.c file, making most of its setup code static.
Document each registers in the meantime.

Its compilation can be later avoided for chips without such registers.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: fix module naming

Since the mv88e6xxx.c file has been renamed, the driver compiled as a
module is called chip.ko instead of mv88e6xxx.ko. Fix this.

Fixes: fad09c73c270 ("net: dsa: mv88e6xxx: rename single-chip support")
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for your net-next
tree.  Most relevant updates are the removal of per-conntrack timers to
use a workqueue/garbage collection approach instead from Florian
Westphal, the hash and numgen expression for nf_tables from Laura
Garcia, updates on nf_tables hash set to honor the NLM_F_EXCL flag,
removal of ip_conntrack sysctl and many other incremental updates on our
Netfilter codebase.

More specifically, they are:

1) Retrieve only 4 bytes to fetch ports in case of non-linear skb
   transport area in dccp, sctp, tcp, udp and udplite protocol
   conntrackers, from Gao Feng.

2) Missing whitespace on error message in physdev match, from Hangbin Liu.

3) Skip redundant IPv4 checksum calculation in nf_dup_ipv4, from Liping Zhang.

4) Add nf_ct_expires() helper function and use it, from Florian Westphal.

5) Replace opencoded nf_ct_kill() call in IPVS conntrack support, also
   from Florian.

6) Rename nf_tables set implementation to nft_set_{name}.c

7) Introduce the hash expression to allow arbitrary hashing of selector
   concatenations, from Laura Garcia Liebana.

8) Remove ip_conntrack sysctl backward compatibility code, this code has
   been around for long time already, and we have two interfaces to do
   this already: nf_conntrack sysctl and ctnetlink.

9) Use nf_conntrack_get_ht() helper function whenever possible, instead
   of opencoding fetch of hashtable pointer and size, patch from Liping Zhang.

10) Add quota expression for nf_tables.

11) Add number generator expression for nf_tables, this supports
    incremental and random generators that can be combined with maps,
    very useful for load balancing purpose, again from Laura Garcia Liebana.

12) Fix a typo in a debug message in FTP conntrack helper, from Colin Ian King.

13) Introduce a nft_chain_parse_hook() helper function to parse chain hook
    configuration, this is used by a follow up patch to perform better chain
    update validation.

14) Add rhashtable_lookup_get_insert_key() to rhashtable and use it from the
    nft_set_hash implementation to honor the NLM_F_EXCL flag.

15) Missing nulls check in nf_conntrack from nf_conntrack_tuple_taken(),
    patch from Florian Westphal.

16) Don't use the DYING bit to know if the conntrack event has been already
    delivered, instead a state variable to track event re-delivery
    states, also from Florian.

17) Remove the per-conntrack timer, use the workqueue approach that was
    discussed during the NFWS, from Florian Westphal.

18) Use the netlink conntrack table dump path to kill stale entries,
    again from Florian.

19) Add a garbage collector to get rid of stale conntracks, from
    Florian.

20) Reschedule garbage collector if eviction rate is high.

21) Get rid of the __nf_ct_kill_acct() helper.

22) Use ARPHRD_ETHER instead of hardcoded 1 from ARP logger.

23) Make nf_log_set() interface assertive on unsupported families.
====================

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

net: ti: cpmac: Fix compiler warning due to type confusion

cpmac_start_xmit() used the max() macro on skb->len (an unsigned int)
and ETH_ZLEN (a signed int literal). This led to the following compiler
warning:

  In file included from include/linux/list.h:8:0,
                   from include/linux/module.h:9,
                   from drivers/net/ethernet/ti/cpmac.c:19:
  drivers/net/ethernet/ti/cpmac.c: In function 'cpmac_start_xmit':
  include/linux/kernel.h:748:17: warning: comparison of distinct pointer
  types lacks a cast
    (void) (&_max1 == &_max2);  \
                   ^
  drivers/net/ethernet/ti/cpmac.c:560:8: note: in expansion of macro 'max'
    len = max(skb->len, ETH_ZLEN);
          ^

On top of this, it assigned the result of the max() macro to a signed
integer whilst all further uses of it result in it being cast to varying
widths of unsigned integer.

Fix this up by using max_t to ensure the comparison is performed as
unsigned integers, and for consistency change the type of the len
variable to unsigned int.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: Add support for ndo_get_vf_config

Adds support for ndo_get_vf_config, also fill the default mac address
that will be provided to the VF by firmware, in case user doesn't
provide one. So user can get the default MAC address address also
through ndo_get_vf_config.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'netns-opt'

Cong Wang says:

====================
net: some minor optimization for netns id

Cong Wang (2):
  vxlan: call peernet2id() in fdb notification
  netns: avoid disabling irq for netns id
====================

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

netns: avoid disabling irq for netns id

We never read or change netns id in hardirq context,
the only place we read netns id in softirq context
is in vxlan_xmit(). So, it should be enough to just
disable BH.

Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

vxlan: call peernet2id() in fdb notification

netns id should be already allocated each time we change
netns, that is, in dev_change_net_namespace() (more precisely
in rtnl_fill_ifinfo()). It is safe to just call peernet2id() here.

Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

openvswitch: Free tmpl with tmpl_free.

When an error occurs during conntrack template creation as part of
actions validation, we need to free the template. Previously we've been
using nf_ct_put() to do this, but nf_ct_tmpl_free() is more appropriate.

Signed-off-by: Joe Stringer <joe@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'liquidio-CN23XX-part-2'

Raghu Vatsavayi says:

====================
liquidio CN23XX support

I am posting the remaining half of patchset after the
acceptance of first half. With this patchset I am able
to completely submit the code of V3 patchset  which you
earlier advised me to split into smaller ones.

This V5 patch also addresses all the comments from previous
submission:
1) Avoid busy loop while reading registers.
2) Other minor comments about debug messages and constants.

Please apply patches in following order as some of the
patches depend on earlier patches.
====================

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

liquidio:CN23XX pause frame support

Adds support for pause frame and priv flag for cn23xx
device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX napi support

This patch adds NAPI related support for cn23xx device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX health monitoring

Adds support for watchdog based health monitoring
of octeon cores on cn23xx device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: ethtool and led control support

This patch adds support for some control operations like
LED identification, ethtool statistics and intr config for
cn23xx device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX octeon3 instruction

Adds support for data path related changes based
on octeon3 instruction header(ih3) for cn23xx.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX IQ access

Adds support for Instruction Queue(IQ) index manipulation
routines through bar1 of cn23xx.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: RX control commands

Adds support for RX control commands on cn23xx device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: link and control commands

This patch adds work queue support for link status and
control commands.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tipc-scalable-broadcast-NACK'

Jon Maloy says:

====================
tipc: improve broadcast NACK mechanism

The broadcast protocol has turned out to not scale well beyond 70-80
nodes, while it is now possible to build TIPC clusters of at least ten
times that size. This commit series improves the NACK/retransmission
mechanism of the broadcast protocol to make is at scalable as the rest
of TIPC.
====================

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

tipc: send broadcast nack directly upon sequence gap detection

Because of the risk of an excessive number of NACK messages and
retransissions, receivers have until now abstained from sending
broadcast NACKS directly upon detection of a packet sequence number
gap. We have instead relied on such gaps being detected by link
protocol STATE message exchange, something that by necessity delays
such detection and subsequent retransmissions.

With the introduction of unicast NACK transmission and rate control
of retransmissions we can now remove this limitation. We now allow
receiving nodes to send NACKS immediately, while coordinating the
permission to do so among the nodes in order to avoid NACK storms.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: rate limit broadcast retransmissions

As cluster sizes grow, so does the amount of identical or overlapping
broadcast NACKs generated by the packet receivers. This often leads to
'NACK crunches' resulting in huge numbers of redundant retransmissions
of the same packet ranges.

In this commit, we introduce rate control of broadcast retransmissions,
so that a retransmitted range cannot be retransmitted again until after
at least 10 ms. This reduces the frequency of duplicate, redundant
retransmissions by an order of magnitude, while having a significant
positive impact on overall throughput and scalability.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tipc: transfer broadcast nacks in link state messages

When we send broadcasts in clusters of more 70-80 nodes, we sometimes
see the broadcast link resetting because of an excessive number of
retransmissions. This is caused by a combination of two factors:

1) A 'NACK crunch", where loss of broadcast packets is discovered
   and NACK'ed by several nodes simultaneously, leading to multiple
   redundant broadcast retransmissions.

2) The fact that the NACKS as such also are sent as broadcast, leading
   to excessive load and packet loss on the transmitting switch/bridge.

This commit deals with the latter problem, by moving sending of
broadcast nacks from the dedicated BCAST_PROTOCOL/NACK message type
to regular unicast LINK_PROTOCOL/STATE messages. We allocate 10 unused
bits in word 8 of the said message for this purpose, and introduce a
new capability bit, TIPC_BCAST_STATE_NACK in order to keep the change
backwards compatible.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: stmmac: dwmac-rk: add pd_gmac support for rk3399

Add the gmac power domain support for rk3399, in order to save more
power consumption.

Signed-off-by: David Wu <david.wu@rock-chips.com>
Signed-off-by: Caesar Wang <wxt@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: stmmac: dwmac-rk: fixes the gmac resume after PD on/off

GMAC Power Domain(PD) will be disabled during suspend.
That will causes GRF registers reset.
So corresponding GRF registers for GMAC must be setup again.

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: Caesar Wang <wxt@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: stmmac: dwmac-rk: add rk3366 & rk3399 specific data

Add constants and callback functions for the dwmac on rk3228/rk3229 socs.
As can be seen, the base structure is the same, only registers and the
bits in them moved slightly.

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: Caesar Wang <wxt@rock-chips.com>
Reviewed-by: Heiko Stuebner <heiko@sntech.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

switchdev: Fix return value of switchdev_port_fdb_dump().

This patch fixes the retun value of switchdev_port_fdb_dump() when
CONFIG_NET_SWITCHDEV is not set. This avoids getting "warning: return makes
integer from pointer without a cast [-Wint-conversion]" when building
when CONFIG_NET_SWITCHDEV is not set under several compiler versions.
This warning is due to commit d297653dd6f07afbe7e6c702a4bcd7615680002e
("rtnetlink: fdb dump: optimize by saving last interface markers").

Signed-off-by: Rami Rosen <rami.rosen@intel.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'bpf-perf-hw-sw-events'

Alexei Starovoitov says:

====================
perf, bpf: add support for bpf in sw/hw perf_events

this patch set is a follow up to the discussion:
https://lkml.kernel.org/r/20160804142853.GO6862%20()%20twins%20!%20programming%20!%20kicks-ass%20!%20net
It turned out to be simpler than what we discussed.

Patches 1-3 is bpf-side prep for the main patch 4
that adds bpf program as an overflow_handler to sw and hw perf_events.

Patches 5 and 6 are examples from myself and Brendan.

Peter,
to implement your suggestion to add ifdef CONFIG_BPF_SYSCALL
inside struct perf_event, I had to shuffle ifdefs in events/core.c
Please double check whether that is what you wanted to see.

v2->v3: fixed few more minor issues
v1->v2: fixed issues spotted by Peter and Daniel.
====================

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

samples/bpf: add sampleip example

sample instruction pointer and frequency count in a BPF map

Signed-off-by: Brendan Gregg <bgregg@netflix.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

samples/bpf: add perf_event+bpf example

The bpf program is called 50 times a second and does hashmap[kern&user_stackid]++
It's primary purpose to check that key bpf helpers like map lookup, update,
get_stackid, trace_printk and ctx access are all working.
It checks:
- PERF_COUNT_HW_CPU_CYCLES on all cpus
- PERF_COUNT_HW_CPU_CYCLES for current process and inherited perf_events to children
- PERF_COUNT_SW_CPU_CLOCK on all cpus
- PERF_COUNT_SW_CPU_CLOCK for current process

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

perf, bpf: add perf events core support for BPF_PROG_TYPE_PERF_EVENT programs

Allow attaching BPF_PROG_TYPE_PERF_EVENT programs to sw and hw perf events
via overflow_handler mechanism.
When program is attached the overflow_handlers become stacked.
The program acts as a filter.
Returning zero from the program means that the normal perf_event_output handler
will not be called and sampling event won't be stored in the ring buffer.

The overflow_handler_context==NULL is an additional safety check
to make sure programs are not attached to hw breakpoints and watchdog
in case other checks (that prevent that now anyway) get accidentally
relaxed in the future.

The program refcnt is incremented in case perf_events are inhereted
when target task is forked.
Similar to kprobe and tracepoint programs there is no ioctl to
detach the program or swap already attached program. The user space
expected to close(perf_event_fd) like it does right now for kprobe+bpf.
That restriction simplifies the code quite a bit.

The invocation of overflow_handler in __perf_event_overflow() is now
done via READ_ONCE, since that pointer can be replaced when the program
is attached while perf_event itself could have been active already.
There is no need to do similar treatment for event->prog, since it's
assigned only once before it's accessed.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

bpf: perf_event progs should only use preallocated maps

Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
preallocated hash maps, since doing memory allocation
in overflow_handler can crash depending on where nmi got triggered.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

bpf: introduce BPF_PROG_TYPE_PERF_EVENT program type

Introduce BPF_PROG_TYPE_PERF_EVENT programs that can be attached to
HW and SW perf events (PERF_TYPE_HARDWARE and PERF_TYPE_SOFTWARE
correspondingly in uapi/linux/perf_event.h)

The program visible context meta structure is
struct bpf_perf_event_data {
    struct pt_regs regs;
     __u64 sample_period;
};
which is accessible directly from the program:
int bpf_prog(struct bpf_perf_event_data *ctx)
{
  ... ctx->sample_period ...
  ... ctx->regs.ip ...
}

The bpf verifier rewrites the accesses into kernel internal
struct bpf_perf_event_data_kern which allows changing
struct perf_sample_data without affecting bpf programs.
New fields can be added to the end of struct bpf_perf_event_data
in the future.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

bpf: support 8-byte metafield access

The verifier supported only 4-byte metafields in
struct __sk_buff and struct xdp_md. The metafields in upcoming
struct bpf_perf_event are 8-byte to match register width in struct pt_regs.
Teach verifier to recognize 8-byte metafield access.
The patch doesn't affect safety of sockets and xdp programs.
They check for 4-byte only ctx access before these conditions are hit.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

mISDN: mark symbols static where possible

We get a few warnings when building kernel with W=1:
drivers/isdn/hardware/mISDN/hfcmulti.c:568:1: warning: no previous declaration for 'enablepcibridge' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:574:1: warning: no previous declaration for 'disablepcibridge' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:580:1: warning: no previous declaration for 'readpcibridge' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:608:1: warning: no previous declaration for 'writepcibridge' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:638:1: warning: no previous declaration for 'cpld_set_reg' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:645:1: warning: no previous declaration for 'cpld_write_reg' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:657:1: warning: no previous declaration for 'cpld_read_reg' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:674:1: warning: no previous declaration for 'vpm_write_address' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:681:1: warning: no previous declaration for 'vpm_read_address' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:695:1: warning: no previous declaration for 'vpm_in' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:716:1: warning: no previous declaration for 'vpm_out' [-Wmissing-declarations]
drivers/isdn/hardware/mISDN/hfcmulti.c:1028:1: warning: no previous declaration for 'plxsd_checksync' [-Wmissing-declarations]
....

In fact, these functions are only used in the file in which they are
declared and don't need a declaration, but can be made static.
so this patch marks these functions with 'static'.

Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: emac: emac gigabit ethernet controller driver

Add support for the Qualcomm Technologies, Inc. EMAC gigabit Ethernet
controller.

This driver supports the following features:
1) Checksum offload.
2) Interrupt coalescing support.
3) SGMII phy.
4) phylib interface for external phy

Based on original work by
Niranjana Vishwanathapura <nvishwan@codeaurora.org>
Gilad Avidov <gavidov@codeaurora.org>

Signed-off-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: remove ds_to_priv

Access the priv member of the dsa_switch structure directly, instead of
having an unnecessary helper.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'br-next'

Nikolay Aleksandrov says:

====================
net: bridge: add per-port unknown multicast flood control

The first patch prepares the forwarding path by having the exact packet
type passed down so we can later filter based on it and the per-port
unknown mcast flood flag introduced in the second patch. It is similar to
how the per-port unknown unicast flood flag works.
Nice side-effects of patch 01 are the slight reduction of tests in the
fast-path and a few minor checkpatch fixes.

v3: don't change br_auto_mask as that will change user-visible behaviour
v2: make pkt_type an enum as per Stephen's comment
====================

Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bridge: add per-port multicast flood flag

Add a per-port flag to control the unknown multicast flood, similar to the
unknown unicast flood flag and break a few long lines in the netlink flag
exports.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bridge: change unicast boolean to exact pkt_type

Remove the unicast flag and introduce an exact pkt_type. That would help us
for the upcoming per-port multicast flood flag and also slightly reduce the
tests in the input fast path.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rtnetlink: fdb dump: optimize by saving last interface markers

fdb dumps spanning multiple skb's currently restart from the first
interface again for every skb. This results in unnecessary
iterations on the already visited interfaces and their fdb
entries. In large scale setups, we have seen this to slow
down fdb dumps considerably. On a system with 30k macs we
see fdb dumps spanning across more than 300 skbs.

To fix the problem, this patch replaces the existing single fdb
marker with three markers: netdev hash entries, netdevs and fdb
index to continue where we left off instead of restarting from the
first netdev. This is consistent with link dumps.

In the process of fixing the performance issue, this patch also
re-implements fix done by
commit 472681d57a5d ("net: ndo_fdb_dump should report -EMSGSIZE to rtnl_fdb_dump")
(with an internal fix from Wilson Kok) in the following ways:
- change ndo_fdb_dump handlers to return error code instead
of the last fdb index
- use cb->args strictly for dump frag markers and not error codes.
This is consistent with other dump functions.

Below results were taken on a system with 1000 netdevs
and 35085 fdb entries:
before patch:
$time bridge fdb show | wc -l
15065

real    1m11.791s
user    0m0.070s
sys 1m8.395s

(existing code does not return all macs)

after patch:
$time bridge fdb show | wc -l
35085

real    0m2.017s
user    0m0.113s
sys 0m1.942s

Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rps: flow_dissector: Add the const for the parameter of flow_keys_have_l4

Add the const for the parameter of flow_keys_have_l4 for the readability.

Signed-off-by: Gao Feng <fgao@ikuai8.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rxrpc: Don't expose skbs to in-kernel users [ver #2]

Don't expose skbs to in-kernel users, such as the AFS filesystem, but
instead provide a notification hook the indicates that a call needs
attention and another that indicates that there's a new call to be
collected.

This makes the following possibilities more achievable:

 (1) Call refcounting can be made simpler if skbs don't hold refs to calls.

 (2) skbs referring to non-data events will be able to be freed much sooner
     rather than being queued for AFS to pick up as rxrpc_kernel_recv_data
     will be able to consult the call state.

 (3) We can shortcut the receive phase when a call is remotely aborted
     because we don't have to go through all the packets to get to the one
     cancelling the operation.

 (4) It makes it easier to do encryption/decryption directly between AFS's
     buffers and sk_buffs.

 (5) Encryption/decryption can more easily be done in the AFS's thread
     contexts - usually that of the userspace process that issued a syscall
     - rather than in one of rxrpc's background threads on a workqueue.

 (6) AFS will be able to wait synchronously on a call inside AF_RXRPC.

To make this work, the following interface function has been added:

     int rxrpc_kernel_recv_data(
struct socket *sock, struct rxrpc_call *call,
void *buffer, size_t bufsize, size_t *_offset,
bool want_more, u32 *_abort_code);

This is the recvmsg equivalent.  It allows the caller to find out about the
state of a specific call and to transfer received data into a buffer
piecemeal.

afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction
logic between them.  They don't wait synchronously yet because the socket
lock needs to be dealt with.

Five interface functions have been removed:

rxrpc_kernel_is_data_last()
     rxrpc_kernel_get_abort_code()
     rxrpc_kernel_get_error_number()
     rxrpc_kernel_free_skb()
     rxrpc_kernel_data_consumed()

As a temporary hack, sk_buffs going to an in-kernel call are queued on the
rxrpc_call struct (->knlrecv_queue) rather than being handed over to the
in-kernel user.  To process the queue internally, a temporary function,
temp_deliver_data() has been added.  This will be replaced with common code
between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a
future patch.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: pegasus: Remove deprecated create_singlethread_workqueue

The workqueue "pegasus_workqueue" queues a single work item per pegasus
instance and hence it doesn't require execution ordering. Hence,
alloc_workqueue has been used to replace the deprecated
create_singlethread_workqueue instance.

The WQ_MEM_RECLAIM flag has been set to ensure forward progress under
memory pressure since it's a network driver.

Since there are fixed number of work items, explicit concurrency
limit is unnecessary here.

Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Petko Manolov <petkan@mip-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bonding: Remove deprecated create_singlethread_workqueue

alloc_ordered_workqueue() with WQ_MEM_RECLAIM set, replaces
deprecated create_singlethread_workqueue(). This is the identity
conversion.

The workqueue "wq" queues multiple work items viz
&bond->mcast_work, &nnw->work, &bond->mii_work, &bond->arp_work,
&bond->alb_work, &bond->mii_work, &bond->ad_work, &bond->slave_arr_work
which require strict execution ordering. Hence, an ordered dedicated
workqueue has been used.

Since, it is a network driver, WQ_MEM_RECLAIM has been set to
ensure forward progress under memory pressure.

Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

sky2: use napi_complete_done

Update the sky2 driver to pass number of packets done to NAPI.
The driver was never updated when napi_complete_done was added.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

l2tp: make nla_policy const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: make nla_policy const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ila: make nla_policy const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

fou: make nla_policy const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

netns: make nla_policy const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

batman: make netlink attributes const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

drop_monitor: make genl_multicast_group const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: make genetlink ctrl ops const

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'stmmac-STM32F429'

Alexandre TORGUE says:

====================
Add Ethernet support on STM32F429

STM32F429 Chip embeds a Synopsys 3.50a MAC IP.

This series enhance current stmmac driver to control it (code already
available) and adds basic glue for STM32F429 chip.

Changes since v5:
 -Fix typo in bindings documentation patch.
 -Change clocks names in stm32-dwmac glue driver / Documentation.
 -After rebase, stm32 ethernet node is now available. It has to be updated
according to new clocks names.

Changes since v4:
 -Fix dirty copy/past in bindings documentation patch.

Changes since v3:
 -Fix "tx-clk" and "rx-clk" as required clocks. Driver and bindings are
modified.

Changes since v2:
 -Fix alphabetic order in Kconfig and Makefile.
 -Improve code according to Joachim review.
 -Binding: remove useless entry.

Changes since v1:
 -Fix Kbuild issue in Kconfig.
 -Remove init/exit callbacks. Suspend/Resume and remove driver is no more
driven in stmmac_pltfr but directly in dwmac-stm32 glue driver.
 -Take into account Joachim review.
====================

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

net: ethernet: stmmac: add support of Synopsys 3.50a MAC IP

Adds support of Synopsys 3.50a MAC IP in stmmac driver.

Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Tested-by: Maxime Coquelin <maxime.coquelin@st.com>
Signed-off-by: Alexandre TORGUE <alexandre.torgue@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Documentation: Bindings: Add STM32 DWMAC glue

Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Alexandre TORGUE <alexandre.torgue@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ethernet: dwmac: add Ethernet glue logic for stm32 chip

stm324xx family chips support Synopsys MAC 3.510 IP.
This patch adds settings for logical glue logic:
-clocks
-mode selection MII or RMII.

Reviewed-by: Joachim Eastwood <manabian@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Tested-by: Maxime Coquelin <maxime.coquelin@st.com>
Signed-off-by: Alexandre TORGUE <alexandre.torgue@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mpls: get rid of trivial returns

return at end of function is useless.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: xgene: fix backward compatibility fix

A bugfix for backward compatibility handling introduced undefined
behavior for the case that of_parse_phandle() does not return
a valid entry, as "gcc -Wmaybe-unused" reports:

drivers/net/ethernet/apm/xgene/xgene_enet_hw.c: In function 'xgene_enet_phy_connect':
drivers/net/ethernet/apm/xgene/xgene_enet_hw.c:776:6: error: 'phy_dev' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/net/ethernet/apm/xgene/xgene_enet_hw.c: In function 'xgene_enet_mdio_config':
drivers/net/ethernet/apm/xgene/xgene_enet_hw.c:776:6: error: 'phy_dev' may be used uninitialized in this function [-Werror=maybe-uninitialized]

We can work around this by removing the check for zero "np", as
of_phy_connect() will correctly handle a NULL argument so we fall
back into the normal error handling case.

Note that I had previously fixed another bug that resulted in the
exact same warning, but this is a different problem that was
introduced after my original fix.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 03377e381bf4 ("drivers: net: xgene: Fix backward compatibility")
Signed-off-by: David S. Miller <davem@davemloft.net>

r8152: fix the coding style with checkpatch.pl

check the coding style with checkpatch.pl and fix the warnings and errors.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'asix-pm-improvements'

Robert Foss says:

====================
net/usb: asix driver improvements

This is a resubmission of v3, since the netdev
mailinlist was not sent the previous submission.

This series improves power management of the asix driver.

 - Suspend/resume support is improved to save needed registers.
 - Device disconnection is improved.
 - Fixes AX88772x resume failures
 - Implementes IEEE 802.3 spec section "22.2.4.1.1 Reset" correctly
 - Fixes AX_CMD_WRITE_MEDIUM_MODE being set incorrectly

Changes since v1:
- Added proper metadata tags to series.
- Added two more patches to series.

Changes since v2:
- Added coverletter
- Tested patches on AX88772A/AX88772B/AX88178/AX88179 hardware
====================

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

net: asix: autoneg will set WRITE_MEDIUM reg

From: Grant Grundler <grundler@chromium.org>

The miii_nway_restart() causes a PHY link change activity and
ax88772_link_reset will be called. link_reset will set
AX_CMD_WRITE_MEDIUM_MODE register correctly.

The asix_write_medium_mode in reset() fills in a default value to the register
which may be different from the negotiation result. So do this first.

Ignore the ret value since it's ignored in XXX_link_reset() functions.

Signed-off-by: Grant Grundler <grundler@google.com>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Tested-by: Robert Foss <robert.foss@collabora.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: asix: see 802.3 spec for phy reset

From: Grant Grundler <grundler@chromium.org>

https://lkml.org/lkml/2014/11/11/947

Ben Hutchings is correct. IEEE 802.3 spec section "22.2.4.1.1 Reset" requires
up to 500ms delay. Mitigate the "max" delay by polling the phy until BCM_RESET
bit is clear.

Signed-off-by: Grant Grundler <grundler@chromium.org>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Tested-by: Robert Foss <robert.foss@collabora.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: asix: Fix AX88772x resume failures

From: Allan Chou <allan@asix.com.tw>

The change fixes AX88772x resume failure by
- Restore incorrect AX88772A PHY registers when resetting
- Need to stop MAC operation when suspending
- Need to restart MII when restoring PHY

Signed-off-by: Allan Chou <allan@asix.com.tw>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Tested-by: Robert Foss <robert.foss@collabora.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: asix: Avoid looping when the device is disconnected

From: Vincent Palatin <vpalatin@chromium.org>

Check the answers from the USB stack and avoid re-sending multiple times
the request if the device has disappeared.

Signed-off-by: Vincent Palatin <vpalatin@chromium.org>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Tested-by: Robert Foss <robert.foss@collabora.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: asix: Add in_pm parameter

From: Freddy Xin <freddy@asix.com.tw>

In order to R/W registers in suspend/resume functions, in_pm flags are
added to some functions to determine whether the nopm version of usb
functions is called.

Save BMCR and ANAR PHY registers in suspend function and restore them
in resume function.

Reset HW in resume function to ensure the PHY works correctly.

Signed-off-by: Freddy Xin <freddy@asix.com.tw>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Tested-by: Robert Foss <robert.foss@collabora.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: axienet: constify ethtool_ops structures

Check for ethtool_ops structures that are only stored in the ethtool_ops
field of a net_device structure or passed as the second argument to
netdev_set_default_ethtool_ops.  These contexts are declared const, so
ethtool_ops structures that have these properties can be declared as const
also.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@r disable optional_qualifier@
identifier i;
position p;
@@
static struct ethtool_ops i@p = { ... };

@ok1@
identifier r.i;
struct net_device e;
position p;
@@
e.ethtool_ops = &i@p;

@ok2@
identifier r.i;
expression e;
position p;
@@
netdev_set_default_ethtool_ops(e, &i@p)

@bad@
position p != {r.p,ok1.p,ok2.p};
identifier r.i;
@@
i@p

@depends on !bad disable optional_qualifier@
identifier r.i;
@@
static
+const
 struct ethtool_ops i = { ... };
// </smpl>

Suggested-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

r8152: constify ethtool_ops structures

Check for ethtool_ops structures that are only stored in the ethtool_ops
field of a net_device structure or passed as the second argument to
netdev_set_default_ethtool_ops.  These contexts are declared const, so
ethtool_ops structures that have these properties can be declared as const
also.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@r disable optional_qualifier@
identifier i;
position p;
@@
static struct ethtool_ops i@p = { ... };

@ok1@
identifier r.i;
struct net_device e;
position p;
@@
e.ethtool_ops = &i@p;

@ok2@
identifier r.i;
expression e;
position p;
@@
netdev_set_default_ethtool_ops(e, &i@p)

@bad@
position p != {r.p,ok1.p,ok2.p};
identifier r.i;
@@
i@p

@depends on !bad disable optional_qualifier@
identifier r.i;
@@
static
+const
 struct ethtool_ops i = { ... };
// </smpl>

Suggested-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: mediatek: constify ethtool_ops structures

Check for ethtool_ops structures that are only stored in the ethtool_ops
field of a net_device structure or passed as the second argument to
netdev_set_default_ethtool_ops.  These contexts are declared const, so
ethtool_ops structures that have these properties can be declared as const
also.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@r disable optional_qualifier@
identifier i;
position p;
@@
static struct ethtool_ops i@p = { ... };

@ok1@
identifier r.i;
struct net_device e;
position p;
@@
e.ethtool_ops = &i@p;

@ok2@
identifier r.i;
expression e;
position p;
@@
netdev_set_default_ethtool_ops(e, &i@p)

@bad@
position p != {r.p,ok1.p,ok2.p};
identifier r.i;
@@
i@p

@depends on !bad disable optional_qualifier@
identifier r.i;
@@
static
+const
 struct ethtool_ops i = { ... };
// </smpl>

Suggested-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'ppp-recursion'

Guillaume Nault says:

====================
ppp: fix deadlock upon recursive xmit

This series fixes the issue reported by Feng where packets looping
through a ppp device makes the module deadlock:
https://marc.info/?l=linux-netdev&m=147134567319038&w=2

The problem can occur on virtual interfaces (e.g. PPP over L2TP, or
PPPoE on vxlan devices), when a PPP packet is routed back to the PPP
interface.

PPP's xmit path isn't reentrant, so patch #1 uses a per-cpu variable
to detect and break recursion. Patch #2 sets the NETIF_F_LLTX flag to
avoid lock inversion issues between ppp and txqueue locks.

There are multiple entry points to the PPP xmit path. This series has
been tested with lockdep and should address recursion issues no matter
how the packet entered the path.

A similar issue in L2TP is not covered by this series:
l2tp_xmit_skb() also isn't reentrant, and it can be called as part of
PPP's xmit path (pppol2tp_xmit()), or directly from the L2TP socket
(l2tp_ppp_sendmsg()). If a packet is sent by l2tp_ppp_sendmsg() and
routed to the parent PPP interface, then it's going to hit
l2tp_xmit_skb() again.

Breaking recursion as done in ppp_generic is not enough, because we'd
still have a lock inversion issue (locking in l2tp_xmit_skb() can
happen before or after locking in ppp_generic). The best approach would
be to use the ip_tunnel functions and remove the socket locking in
l2tp_xmit_skb(). But that'd be something for net-next.
====================

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

ppp: declare PPP devices as LLTX

ppp_xmit_process() already locks the xmit path. If HARD_TX_LOCK() tries
to hold the _xmit_lock we can get lock inversion.

[  973.726130] ======================================================
[  973.727311] [ INFO: possible circular locking dependency detected ]
[  973.728546] 4.8.0-rc2 #1 Tainted: G           O
[  973.728986] -------------------------------------------------------
[  973.728986] accel-pppd/1806 is trying to acquire lock:
[  973.728986]  (&qdisc_xmit_lock_key){+.-...}, at: [<ffffffff8146f6fe>] sch_direct_xmit+0x8d/0x221
[  973.728986]
[  973.728986] but task is already holding lock:
[  973.728986]  (l2tp_sock){+.-...}, at: [<ffffffffa0202c4a>] l2tp_xmit_skb+0x1e8/0x5d7 [l2tp_core]
[  973.728986]
[  973.728986] which lock already depends on the new lock.
[  973.728986]
[  973.728986]
[  973.728986] the existing dependency chain (in reverse order) is:
[  973.728986]
-> #3 (l2tp_sock){+.-...}:
[  973.728986]        [<ffffffff810b3130>] lock_acquire+0x150/0x217
[  973.728986]        [<ffffffff815752f4>] _raw_spin_lock+0x2d/0x3c
[  973.728986]        [<ffffffffa0202c4a>] l2tp_xmit_skb+0x1e8/0x5d7 [l2tp_core]
[  973.728986]        [<ffffffffa01b2466>] pppol2tp_xmit+0x1f2/0x25e [l2tp_ppp]
[  973.728986]        [<ffffffffa0184f59>] ppp_channel_push+0xb5/0x14a [ppp_generic]
[  973.728986]        [<ffffffffa01853ed>] ppp_write+0x104/0x11c [ppp_generic]
[  973.728986]        [<ffffffff811b2ec6>] __vfs_write+0x56/0x120
[  973.728986]        [<ffffffff811b3f4c>] vfs_write+0xbd/0x11b
[  973.728986]        [<ffffffff811b4cb2>] SyS_write+0x5e/0x96
[  973.728986]        [<ffffffff81575ba5>] entry_SYSCALL_64_fastpath+0x18/0xa8
[  973.728986]
-> #2 (&(&pch->downl)->rlock){+.-...}:
[  973.728986]        [<ffffffff810b3130>] lock_acquire+0x150/0x217
[  973.728986]        [<ffffffff81575334>] _raw_spin_lock_bh+0x31/0x40
[  973.728986]        [<ffffffffa01808e2>] ppp_push+0xa7/0x82d [ppp_generic]
[  973.728986]        [<ffffffffa0184675>] __ppp_xmit_process+0x48/0x877 [ppp_generic]
[  973.728986]        [<ffffffffa018505b>] ppp_xmit_process+0x4b/0xaf [ppp_generic]
[  973.728986]        [<ffffffffa01853f7>] ppp_write+0x10e/0x11c [ppp_generic]
[  973.728986]        [<ffffffff811b2ec6>] __vfs_write+0x56/0x120
[  973.728986]        [<ffffffff811b3f4c>] vfs_write+0xbd/0x11b
[  973.728986]        [<ffffffff811b4cb2>] SyS_write+0x5e/0x96
[  973.728986]        [<ffffffff81575ba5>] entry_SYSCALL_64_fastpath+0x18/0xa8
[  973.728986]
-> #1 (&(&ppp->wlock)->rlock){+.-...}:
[  973.728986]        [<ffffffff810b3130>] lock_acquire+0x150/0x217
[  973.728986]        [<ffffffff81575334>] _raw_spin_lock_bh+0x31/0x40
[  973.728986]        [<ffffffffa0184654>] __ppp_xmit_process+0x27/0x877 [ppp_generic]
[  973.728986]        [<ffffffffa018505b>] ppp_xmit_process+0x4b/0xaf [ppp_generic]
[  973.728986]        [<ffffffffa01852da>] ppp_start_xmit+0x21b/0x22a [ppp_generic]
[  973.728986]        [<ffffffff8143f767>] dev_hard_start_xmit+0x1a9/0x43d
[  973.728986]        [<ffffffff8146f747>] sch_direct_xmit+0xd6/0x221
[  973.728986]        [<ffffffff814401e4>] __dev_queue_xmit+0x62a/0x912
[  973.728986]        [<ffffffff814404d7>] dev_queue_xmit+0xb/0xd
[  973.728986]        [<ffffffff81449978>] neigh_direct_output+0xc/0xe
[  973.728986]        [<ffffffff8150e62b>] ip6_finish_output2+0x5a9/0x623
[  973.728986]        [<ffffffff81512128>] ip6_output+0x15e/0x16a
[  973.728986]        [<ffffffff8153ef86>] dst_output+0x76/0x7f
[  973.728986]        [<ffffffff8153f737>] mld_sendpack+0x335/0x404
[  973.728986]        [<ffffffff81541c61>] mld_send_initial_cr.part.21+0x99/0xa2
[  973.728986]        [<ffffffff8154441d>] ipv6_mc_dad_complete+0x42/0x71
[  973.728986]        [<ffffffff8151c4bd>] addrconf_dad_completed+0x1cf/0x2ea
[  973.728986]        [<ffffffff8151e4fa>] addrconf_dad_work+0x453/0x520
[  973.728986]        [<ffffffff8107a393>] process_one_work+0x365/0x6f0
[  973.728986]        [<ffffffff8107aecd>] worker_thread+0x2de/0x421
[  973.728986]        [<ffffffff810816fb>] kthread+0x121/0x130
[  973.728986]        [<ffffffff81575dbf>] ret_from_fork+0x1f/0x40
[  973.728986]
-> #0 (&qdisc_xmit_lock_key){+.-...}:
[  973.728986]        [<ffffffff810b28d6>] __lock_acquire+0x1118/0x1483
[  973.728986]        [<ffffffff810b3130>] lock_acquire+0x150/0x217
[  973.728986]        [<ffffffff815752f4>] _raw_spin_lock+0x2d/0x3c
[  973.728986]        [<ffffffff8146f6fe>] sch_direct_xmit+0x8d/0x221
[  973.728986]        [<ffffffff814401e4>] __dev_queue_xmit+0x62a/0x912
[  973.728986]        [<ffffffff814404d7>] dev_queue_xmit+0xb/0xd
[  973.728986]        [<ffffffff81449978>] neigh_direct_output+0xc/0xe
[  973.728986]        [<ffffffff81487811>] ip_finish_output2+0x5db/0x609
[  973.728986]        [<ffffffff81489590>] ip_finish_output+0x152/0x15e
[  973.728986]        [<ffffffff8148a0d4>] ip_output+0x8c/0x96
[  973.728986]        [<ffffffff81489652>] ip_local_out+0x41/0x4a
[  973.728986]        [<ffffffff81489e7d>] ip_queue_xmit+0x5a5/0x609
[  973.728986]        [<ffffffffa0202fe4>] l2tp_xmit_skb+0x582/0x5d7 [l2tp_core]
[  973.728986]        [<ffffffffa01b2466>] pppol2tp_xmit+0x1f2/0x25e [l2tp_ppp]
[  973.728986]        [<ffffffffa0184f59>] ppp_channel_push+0xb5/0x14a [ppp_generic]
[  973.728986]        [<ffffffffa01853ed>] ppp_write+0x104/0x11c [ppp_generic]
[  973.728986]        [<ffffffff811b2ec6>] __vfs_write+0x56/0x120
[  973.728986]        [<ffffffff811b3f4c>] vfs_write+0xbd/0x11b
[  973.728986]        [<ffffffff811b4cb2>] SyS_write+0x5e/0x96
[  973.728986]        [<ffffffff81575ba5>] entry_SYSCALL_64_fastpath+0x18/0xa8
[  973.728986]
[  973.728986] other info that might help us debug this:
[  973.728986]
[  973.728986] Chain exists of:
  &qdisc_xmit_lock_key --> &(&pch->downl)->rlock --> l2tp_sock

[  973.728986]  Possible unsafe locking scenario:
[  973.728986]
[  973.728986]        CPU0                    CPU1
[  973.728986]        ----                    ----
[  973.728986]   lock(l2tp_sock);
[  973.728986]                                lock(&(&pch->downl)->rlock);
[  973.728986]                                lock(l2tp_sock);
[  973.728986]   lock(&qdisc_xmit_lock_key);
[  973.728986]
[  973.728986]  *** DEADLOCK ***
[  973.728986]
[  973.728986] 6 locks held by accel-pppd/1806:
[  973.728986]  #0:  (&(&pch->downl)->rlock){+.-...}, at: [<ffffffffa0184efa>] ppp_channel_push+0x56/0x14a [ppp_generic]
[  973.728986]  #1:  (l2tp_sock){+.-...}, at: [<ffffffffa0202c4a>] l2tp_xmit_skb+0x1e8/0x5d7 [l2tp_core]
[  973.728986]  #2:  (rcu_read_lock){......}, at: [<ffffffff81486981>] rcu_lock_acquire+0x0/0x20
[  973.728986]  #3:  (rcu_read_lock_bh){......}, at: [<ffffffff81486981>] rcu_lock_acquire+0x0/0x20
[  973.728986]  #4:  (rcu_read_lock_bh){......}, at: [<ffffffff814340e3>] rcu_lock_acquire+0x0/0x20
[  973.728986]  #5:  (dev->qdisc_running_key ?: &qdisc_running_key#2){+.....}, at: [<ffffffff8144011e>] __dev_queue_xmit+0x564/0x912
[  973.728986]
[  973.728986] stack backtrace:
[  973.728986] CPU: 2 PID: 1806 Comm: accel-pppd Tainted: G           O    4.8.0-rc2 #1
[  973.728986] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
[  973.728986]  ffff7fffffffffff ffff88003436f850 ffffffff812a20f4 ffffffff82156e30
[  973.728986]  ffffffff82156920 ffff88003436f890 ffffffff8115c759 ffff88003344ae00
[  973.728986]  ffff88003344b5c0 0000000000000002 0000000000000006 ffff88003344b5e8
[  973.728986] Call Trace:
[  973.728986]  [<ffffffff812a20f4>] dump_stack+0x67/0x90
[  973.728986]  [<ffffffff8115c759>] print_circular_bug+0x22e/0x23c
[  973.728986]  [<ffffffff810b28d6>] __lock_acquire+0x1118/0x1483
[  973.728986]  [<ffffffff810b3130>] lock_acquire+0x150/0x217
[  973.728986]  [<ffffffff810b3130>] ? lock_acquire+0x150/0x217
[  973.728986]  [<ffffffff8146f6fe>] ? sch_direct_xmit+0x8d/0x221
[  973.728986]  [<ffffffff815752f4>] _raw_spin_lock+0x2d/0x3c
[  973.728986]  [<ffffffff8146f6fe>] ? sch_direct_xmit+0x8d/0x221
[  973.728986]  [<ffffffff8146f6fe>] sch_direct_xmit+0x8d/0x221
[  973.728986]  [<ffffffff814401e4>] __dev_queue_xmit+0x62a/0x912
[  973.728986]  [<ffffffff814404d7>] dev_queue_xmit+0xb/0xd
[  973.728986]  [<ffffffff81449978>] neigh_direct_output+0xc/0xe
[  973.728986]  [<ffffffff81487811>] ip_finish_output2+0x5db/0x609
[  973.728986]  [<ffffffff81486853>] ? dst_mtu+0x29/0x2e
[  973.728986]  [<ffffffff81489590>] ip_finish_output+0x152/0x15e
[  973.728986]  [<ffffffff8148a0bc>] ? ip_output+0x74/0x96
[  973.728986]  [<ffffffff8148a0d4>] ip_output+0x8c/0x96
[  973.728986]  [<ffffffff81489652>] ip_local_out+0x41/0x4a
[  973.728986]  [<ffffffff81489e7d>] ip_queue_xmit+0x5a5/0x609
[  973.728986]  [<ffffffff814c559e>] ? udp_set_csum+0x207/0x21e
[  973.728986]  [<ffffffffa0202fe4>] l2tp_xmit_skb+0x582/0x5d7 [l2tp_core]
[  973.728986]  [<ffffffffa01b2466>] pppol2tp_xmit+0x1f2/0x25e [l2tp_ppp]
[  973.728986]  [<ffffffffa0184f59>] ppp_channel_push+0xb5/0x14a [ppp_generic]
[  973.728986]  [<ffffffffa01853ed>] ppp_write+0x104/0x11c [ppp_generic]
[  973.728986]  [<ffffffff811b2ec6>] __vfs_write+0x56/0x120
[  973.728986]  [<ffffffff8124c11d>] ? fsnotify_perm+0x27/0x95
[  973.728986]  [<ffffffff8124d41d>] ? security_file_permission+0x4d/0x54
[  973.728986]  [<ffffffff811b3f4c>] vfs_write+0xbd/0x11b
[  973.728986]  [<ffffffff811b4cb2>] SyS_write+0x5e/0x96
[  973.728986]  [<ffffffff81575ba5>] entry_SYSCALL_64_fastpath+0x18/0xa8
[  973.728986]  [<ffffffff810ae0fa>] ? trace_hardirqs_off_caller+0x121/0x12f

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

ppp: avoid dealock on recursive xmit

In case of misconfiguration, a virtual PPP channel might send packets
back to their parent PPP interface. This typically happens in
misconfigured L2TP setups, where PPP's peer IP address is set with the
IP of the L2TP peer.
When that happens the system hangs due to PPP trying to recursively
lock its xmit path.

[  243.332155] BUG: spinlock recursion on CPU#1, accel-pppd/926
[  243.333272]  lock: 0xffff880033d90f18, .magic: dead4ead, .owner: accel-pppd/926, .owner_cpu: 1
[  243.334859] CPU: 1 PID: 926 Comm: accel-pppd Not tainted 4.8.0-rc2 #1
[  243.336010] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
[  243.336018]  ffff7fffffffffff ffff8800319a77a0 ffffffff8128de85 ffff880033d90f18
[  243.336018]  ffff880033ad8000 ffff8800319a77d8 ffffffff810ad7c0 ffffffff0000039e
[  243.336018]  ffff880033d90f18 ffff880033d90f60 ffff880033d90f18 ffff880033d90f28
[  243.336018] Call Trace:
[  243.336018]  [<ffffffff8128de85>] dump_stack+0x4f/0x65
[  243.336018]  [<ffffffff810ad7c0>] spin_dump+0xe1/0xeb
[  243.336018]  [<ffffffff810ad7f0>] spin_bug+0x26/0x28
[  243.336018]  [<ffffffff810ad8b9>] do_raw_spin_lock+0x5c/0x160
[  243.336018]  [<ffffffff815522aa>] _raw_spin_lock_bh+0x35/0x3c
[  243.336018]  [<ffffffffa01a88e2>] ? ppp_push+0xa7/0x82d [ppp_generic]
[  243.336018]  [<ffffffffa01a88e2>] ppp_push+0xa7/0x82d [ppp_generic]
[  243.336018]  [<ffffffff810adada>] ? do_raw_spin_unlock+0xc2/0xcc
[  243.336018]  [<ffffffff81084962>] ? preempt_count_sub+0x13/0xc7
[  243.336018]  [<ffffffff81552438>] ? _raw_spin_unlock_irqrestore+0x34/0x49
[  243.336018]  [<ffffffffa01ac657>] ppp_xmit_process+0x48/0x877 [ppp_generic]
[  243.336018]  [<ffffffff81084962>] ? preempt_count_sub+0x13/0xc7
[  243.336018]  [<ffffffff81408cd3>] ? skb_queue_tail+0x71/0x7c
[  243.336018]  [<ffffffffa01ad1c5>] ppp_start_xmit+0x21b/0x22a [ppp_generic]
[  243.336018]  [<ffffffff81426af1>] dev_hard_start_xmit+0x15e/0x32c
[  243.336018]  [<ffffffff81454ed7>] sch_direct_xmit+0xd6/0x221
[  243.336018]  [<ffffffff814273a8>] __dev_queue_xmit+0x52a/0x820
[  243.336018]  [<ffffffff814276a9>] dev_queue_xmit+0xb/0xd
[  243.336018]  [<ffffffff81430a3c>] neigh_direct_output+0xc/0xe
[  243.336018]  [<ffffffff8146b5d7>] ip_finish_output2+0x4d2/0x548
[  243.336018]  [<ffffffff8146a8e6>] ? dst_mtu+0x29/0x2e
[  243.336018]  [<ffffffff8146d49c>] ip_finish_output+0x152/0x15e
[  243.336018]  [<ffffffff8146df84>] ? ip_output+0x74/0x96
[  243.336018]  [<ffffffff8146df9c>] ip_output+0x8c/0x96
[  243.336018]  [<ffffffff8146d55e>] ip_local_out+0x41/0x4a
[  243.336018]  [<ffffffff8146dd15>] ip_queue_xmit+0x531/0x5c5
[  243.336018]  [<ffffffff814a82cd>] ? udp_set_csum+0x207/0x21e
[  243.336018]  [<ffffffffa01f2f04>] l2tp_xmit_skb+0x582/0x5d7 [l2tp_core]
[  243.336018]  [<ffffffffa01ea458>] pppol2tp_xmit+0x1eb/0x257 [l2tp_ppp]
[  243.336018]  [<ffffffffa01acf17>] ppp_channel_push+0x91/0x102 [ppp_generic]
[  243.336018]  [<ffffffffa01ad2d8>] ppp_write+0x104/0x11c [ppp_generic]
[  243.336018]  [<ffffffff811a3c1e>] __vfs_write+0x56/0x120
[  243.336018]  [<ffffffff81239801>] ? fsnotify_perm+0x27/0x95
[  243.336018]  [<ffffffff8123ab01>] ? security_file_permission+0x4d/0x54
[  243.336018]  [<ffffffff811a4ca4>] vfs_write+0xbd/0x11b
[  243.336018]  [<ffffffff811a5a0a>] SyS_write+0x5e/0x96
[  243.336018]  [<ffffffff81552a1b>] entry_SYSCALL_64_fastpath+0x13/0x94

The main entry points for sending packets over a PPP unit are the
.write() and .ndo_start_xmit() callbacks (simplified view):

.write(unit fd) or .ndo_start_xmit()
       \
        CALL ppp_xmit_process()
               \
                LOCK unit's xmit path (ppp->wlock)
                |
                CALL ppp_push()
                       \
                        LOCK channel's xmit path (chan->downl)
                        |
                        CALL lower layer's .start_xmit() callback
                               \
                                ... might recursively call .ndo_start_xmit() ...
                               /
                        RETURN from .start_xmit()
                        |
                        UNLOCK channel's xmit path
                       /
                RETURN from ppp_push()
                |
                UNLOCK unit's xmit path
               /
        RETURN from ppp_xmit_process()

Packets can also be directly sent on channels (e.g. LCP packets):

.write(channel fd) or ppp_output_wakeup()
       \
        CALL ppp_channel_push()
               \
                LOCK channel's xmit path (chan->downl)
                |
                CALL lower layer's .start_xmit() callback
                       \
                        ... might call .ndo_start_xmit() ...
                       /
                RETURN from .start_xmit()
                |
                UNLOCK channel's xmit path
               /
        RETURN from ppp_channel_push()

Key points about the lower layer's .start_xmit() callback:

  * It can be called directly by a channel fd .write() or by
    ppp_output_wakeup() or indirectly by a unit fd .write() or by
    .ndo_start_xmit().

  * In any case, it's always called with chan->downl held.

  * It might route the packet back to its parent unit using
    .ndo_start_xmit() as entry point.

This patch detects and breaks recursion in ppp_xmit_process(). This
function is a good candidate for the task because it's called early
enough after .ndo_start_xmit(), it's always part of the recursion
loop and it's on the path of whatever entry point is used to send
a packet on a PPP unit.

Recursion detection is done using the per-cpu ppp_xmit_recursion
variable.

Since ppp_channel_push() too locks the channel's xmit path and calls
the lower layer's .start_xmit() callback, we need to also increment
ppp_xmit_recursion there. However there's no need to check for
recursion, as it's out of the recursion loop.

Reported-by: Feng Gao <gfree.wind@gmail.com>
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

xgbe: constify get_netdev_ops and get_ethtool_ops

Casting away const is bad practice. Since this is ARM specific driver
don't have hardware actually test this.

Having getter functions for ops is really unnecessary code bloat, but
not going to touch that.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'dsa-mdb-support'

Vivien Didelot says:

====================
net: dsa: add MDB support

This patchset adds the switchdev MDB object support to the DSA layer.

The MDB support for the mv88e6xxx driver is very similar to the FDB
support. The FDB operations care about unicast addresses while the MDB
operations care about multicast addresses.

Both operation set load/purge/dump the Address Translation Table (ATU),
thus common code is used.

Changes in v2 based on Andrew's comments:
  - drop "group" in multicast database related doc and comment
  - change _one for more relevant _fid in mv88e6xxx_port_db_dump_one
  - return -EOPNOTSUPP if switchdev obj ID is neither _FDB nor _MDB
====================

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: add MDB support

Add support for the MDB operations. This consists of
loading/purging/dumping multicast addresses for a given port in the ATU.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: mv88e6xxx: make switchdev DB ops generic

The MDB support for the mv88e6xxx driver will be very similar to the FDB
support, since it consists of loading/purging/dumping address to/from
the Address Translation Unit (ATU).

Prepare the support for MDB by making the FDB code accessing the ATU
generic. The FDB operations now provide access to the unicast addresses
while the MDB operations will provide access to the multicast addresses.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: dsa: add MDB support

Add SWITCHDEV_OBJ_ID_PORT_MDB support to the DSA layer.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'liquidio-CN23XX-part-1'

Raghu Vatsavayi says:

====================
liquidio CN23XX support

Following patchset adds support for new device "CN23XX" in
liquidio family of adapters. As adviced by you I have split
the previous V3 patch of 18 patches into two halves. This
first patchset has first 10 patches, which are tested against
net-next. I will post the second half after this one.

This V4 patch also addressed all the comments from previous
submission:
1) Avoid busy loop while reading registers.
2) Other minor comments about debug messages and constants.

Please apply patches in following order as some of the
patches depend on earlier patches.
====================

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

liquidio: CN23XX firmware download

Add firmware download support for cn23xx device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: MSIX support for CN23XX

This patch adds support msix interrupt for cn23xx device.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX queue manipulation

This patch adds support for cn23xx queue manipulation.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX register setup

Adds support for initializing cn23xx device registers
related to mac, input/output and pf global config.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX device init and sriov config

Add support for cn23xx device init and sriov queue config.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX queue definitions

Add support for cn23xx specific queue definitions and
features.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: CN23XX register definitions

This patch adds register definitions and structures for new
device cn23xx.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: Common enable irq function

Add support of common irq enable functionality for both
iq(instruction queue) and oq(output queue).

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

liquidio: Firmware version management

This patch contains changes for firmware version management.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>