When running consumer and/or producer operations and empty checks in
parallel its possible to have the empty check run past the end of the
array. The scenario occurs when an empty check is run while
__ptr_ring_discard_one() is in progress. Specifically after the
consumer_head is incremented but before (consumer_head >= ring_size)
check is made and the consumer head is zeroe'd.

To resolve this, without having to rework how consumer/producer ops
work on the array, simply add an extra dummy slot to the end of the
array. Even if we did a rework to avoid the extra slot it looks
like the normal case checks would suffer some so best to just
allocate an extra pointer.

Reported-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Fixes: c5ad119f

 ("net: sched: pfifo_fast use skb_array")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv6/ip6_gre.c is a case of parallel adds.

include/trace/events/tcp.h is a little bit more tricky.  The removal
of in-trace-macro ifdefs in 'net' paralleled with moving
show_tcp_state_name and friends over to include/trace/events/sock.h
in 'net-next'.

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

Pull networking fixes from David Miller:

 1) IPv6 gre tunnels end up with different default features enabled
    depending upon whether netlink or ioctls are used to bring them up.
    Fix from Alexey Kodanev.

 2) Fix read past end of user control message in RDS< from Avinash
    Repaka.

 3) Missing RCU barrier in mini qdisc code, from Cong Wang.

 4) Missing policy put when reusing per-cpu route entries, from Florian
    Westphal.

 5) Handle nested PCI errors properly in bnx2x driver, from Guilherme G.
    Piccoli.

 6) Run nested transport mode IPSEC packets via tasklet, from Herbert
    Xu.

 7) Fix handling poll() for stream sockets in tipc, from Parthasarathy
    Bhuvaragan.

 8) Fix two stack-out-of-bounds issues in IPSEC, from Steffen Klassert.

 9) Another zerocopy ubuf handling fix, from Willem de Bruijn.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (33 commits)
  strparser: Call sock_owned_by_user_nocheck
  sock: Ad...

Pull drm fixes from Dave Airlie:
 "nouveau and i915 regression fixes"

* tag 'drm-fixes-for-v4.15-rc6' of git://people.freedesktop.org/~airlied/linux:
  drm/nouveau: fix race when adding delayed work items
  i915: Reject CCS modifiers for pipe C on Geminilake
  drm/i915/gvt: Fix pipe A enable as default for vgpu

Pull clk fix from Stephen Boyd:
 "One more fix for the runtime PM clk patches. We're calling a runtime
  PM API that may schedule from somewhere that we can't do that. We
  change to the async version of pm_runtime_put() to fix it"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  clk: use atomic runtime pm api in clk_core_is_enabled

Merge tag 'led_fixes_for_4.15-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/j.anaszewski/linux-leds

Pull LED fix from Jacek Anaszewski:
 "A single LED fix for brightness setting when delay_off is 0"

* tag 'led_fixes_for_4.15-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/j.anaszewski/linux-leds:
  led: core: Fix brightness setting when setting delay_off=0

Pull rdma fixes from Jason Gunthorpe:
 "This is the next batch of for-rc patches from RDMA. It includes the
  fix for the ipoib regression I mentioned last time, and the result of
  a fairly major debugging effort to get iser working reliably on cxgb4
  hardware - it turns out the cxgb4 driver was not handling QP error
  flushing properly causing iser to fail.

   - cxgb4 fix for an iser testing failure as debugged by Steve and
     Sagi. The problem was a driver bug in the handling of shutting down
     a QP.

   - Various vmw_pvrdma fixes for bogus WARN_ON, missed resource free on
     error unwind and a use after free bug

   - Improper congestion counter values on mlx5 when link aggregation is
     enabled

   - ipoib lockdep regression introduced in this merge window

   - hfi1 regression supporting the device in a VM introduced in a
     recent patch

   - Typo that breaks future uAPI compatibility in the verbs core

   - More SELinux related oops fixing

   - Fix an oops during error unwind in mlx5"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  IB/mlx5: Fix mlx5_ib_alloc_mr error flow
  IB/core: Verify that QP is security enabled in create and destroy
  IB/uverbs: Fix command checking as part of ib_uverbs_ex_modify_qp()
  IB/mlx5: Serialize access to the VMA list
  IB/hfi: Only read capability registers if the capability exists
  IB/ipoib: Fix lockdep issue found on ipoib_ib_dev_heavy_flush
  IB/mlx5: Fix congestion counters in LAG mode
  RDMA/vmw_pvrdma: Avoid use after free due to QP/CQ/SRQ destroy
  RDMA/vmw_pvrdma: Use refcount_dec_and_test to avoid warning
  RDMA/vmw_pvrdma: Call ib_umem_release on destroy QP path
  iw_cxgb4: when flushing, complete all wrs in a chain
  iw_cxgb4: reflect the original WR opcode in drain cqes
  iw_cxgb4: Only validate the MSN for successful completions

Saeed Mahameed says:

====================
Mellanox, mlx5 E-Switch updates 2017-12-19

This series includes updates for mlx5 E-Switch infrastructures,
to be merged into net-next and rdma-next trees.

Mark's patches provide E-Switch refactoring that generalize the mlx5
E-Switch vf representors interfaces and data structures. The serious is
mainly focused on moving ethernet (netdev) specific representors logic out
of E-Switch (eswitch.c) into mlx5e representor module (en_rep.c), which
provides better separation and allows future support for other types of vf
representors (e.g. RDMA).

Gal's patches at the end of this serious, provide a simple syntax fix and
two other patches that handles vport ingress/egress ACL steering name
spaces to be aligned with the Firmware/Hardware specs.

V1->V2:
 - Addressed coding style comments in patches #1 and #7
 - The series is still based on rc4, as now I see net-next is also @rc4.

V2->V3:
 - Fixed compilation warning, reported by Dave.

Please pull and let me know if there's any problem.
====================

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

Each vport has its own root flow table for the ACL flow tables and root
flow table is per namespace, therefore we should create a namespace for
each vport.

Fixes: efdc810b

 ("net/mlx5: Flow steering, Add vport ACL support")
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

The functions names do not represent their actions, switch the mistaken
ingress/egress naming.

Fixes: fba53f7b

 ("net/mlx5: Introduce mlx5_flow_steering structure")
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Using the address of a static array is the same as using its name (in
this specific use-case), but it's confusing and makes the code less
readable.

Fixes: 1bd27b11 ("net/mlx5: Introduce E-switch QoS management")
Fixes: bd77bf1c

 ("net/mlx5: Add SRIOV VF max rate configuration support")
Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Move struct mlx5_esw_sq which keeps send-to-vport rule to from the eswitch
code to mlx5e and rename it to better reflect where it belongs

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Now that we don't store type dependent data in struct mlx5_eswitch_rep
we can create a generic interface, and representor type.

struct mlx5_eswitch_rep will store an array of interfaces, each
interface is used by a different representor type.

Once we moved to a more generic interface, rdma driver representors can
be added and utilize the same mechanism as the Ethernet driver
representors use.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Tom Herbert says:

====================
strparser: Fix lockdep issue

When sock_owned_by_user returns true in strparser. Fix is to add and
call sock_owned_by_user_nocheck since the check for owned by user is
not an error condition in this case.
====================

Fixes: 43a0c675

 ("strparser: Stream parser for messages")
Reported-by: syzbot <syzkaller@googlegroups.com>
Reported-and-tested-by:  <syzbot+c91c53af67f9ebe599a337d2e70950366153b295@syzkaller.appspotmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

strparser wants to check socket ownership without producing any
warnings. As indicated by the comment in the code, it is permissible
for owned_by_user to return true.

Fixes: 43a0c675

 ("strparser: Stream parser for messages")
Reported-by: syzbot <syzkaller@googlegroups.com>
Reported-and-tested-by:  <syzbot+c91c53af67f9ebe599a337d2e70950366153b295@syzkaller.appspotmail.com>
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

This allows checking socket lock ownership with producing lockdep
warnings.

Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

skb_copy_ubufs must unclone before it is safe to modify its
skb_shared_info with skb_zcopy_clear.

Commit b90ddd56 ("skbuff: skb_copy_ubufs must release uarg even
without user frags") ensures that all skbs release their zerocopy
state, even those without frags.

But I forgot an edge case where such an skb arrives that is cloned.

The stack does not build such packets. Vhost/tun skbs have their
frags orphaned before cloning. TCP skbs only attach zerocopy state
when a frag is added.

But if TCP packets can be trimmed or linearized, this might occur.
Tracing the code I found no instance so far (e.g., skb_linearize
ends up calling skb_zcopy_clear if !skb->data_len).

Still, it is non-obvious that no path exists. And it is fragile to
rely on this.

Fixes: b90ddd56

 ("skbuff: skb_copy_ubufs must release uarg even without user frags")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Tariq Toukan says:

====================
mlx4 misc for 4.16

This patchset contains misc cleanups and improvements
to the mlx4 Core and Eth drivers.

In patches 1 and 2 I reduce and reorder the branches in the RX csum flow.
In patch 3 I align the FMR unmapping flow with the device spec, to allow
  a remapping afterwards.
Patch 4 by Moni changes the default QoS settings so that a pause
  frame stops all traffic regardless of its prio.

Series generated against net-next commit:
836df24a

 net: hns3: hns3_get_channels() can be static
====================

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

Change the default mapping between TC and TCG as follows:

Prio     |             TC/TCG
         |      from             to
         |    (set by FW)      (set by SW)
---------+-----------------------------------
0        |      0/0              0/7
1        |      1/0              0/6
2        |      2/0              0/5
3        |      3/0              0/4
4        |      4/0              0/3
5        |      5/0              0/2
6        |      6/0              0/1
7        |      7/0              0/0

These new settings cause that a pause frame for any prio stops
traffic for all prios.

Fixes: 564c274c

 ("net/mlx4_en: DCB QoS support")
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Remove redundant and not essential operations in fmr unmap/free.
According to device spec, in FMR unmap it is sufficient to set
ownership bit to SW. This allows remapping afterwards.

Fixes: 8ad11fb6

 ("IB/mlx4: Implement FMRs")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Use early goto commands, and save else branches.
This uses less indentations and brackets, making the code
more readable.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Do not check IPv6 bit in cqe status if CONFIG_IPV6 is not enabled.
Function check_csum() is reached only with IPv4 or IPv6 set (if enabled),
if IPv6 is not set (or is not enabled) it is redundant to test the
IPv4 bit.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If the qdisc is not found here, it is going to be created. Therefore,
this is not an error path. Remove the extack message set and don't
confuse user with error message in case the qdisc was created
successfully.

Fixes: 09215598

 ("net: sched: sch_api: handle generic qdisc errors")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In commit 42b531de ("tipc: Fix missing connection request
handling"), we replaced unconditional wakeup() with condtional
wakeup for clients with flags POLLIN | POLLRDNORM | POLLRDBAND.

This breaks the applications which do a connect followed by poll
with POLLOUT flag. These applications are not woken when the
connection is ESTABLISHED and hence sleep forever.

In this commit, we fix it by including the POLLOUT event for
sockets in TIPC_CONNECTING state.

Fixes: 42b531de

 ("tipc: Fix missing connection request handling")
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Kunihiko Hayashi says:

====================
net: add UniPhier AVE ethernet support

This series adds support for Socionext AVE ethernet controller implemented
on UniPhier SoCs. This driver supports RGMII/RMII modes.

v8: https://www.spinics.net/lists/netdev/msg474374.html

The PHY patch included in v1 has already separated in:
http://www.spinics.net/lists/netdev/msg454595.html

Changes since v8:
- move operators at the beginning of the line to the end of the previous line
- dt-bindings: add blank lines before mdio and phy subnodes

Changes since v7:
- dt-bindings: fix mdio subnode description

Changes since v6:
- sort the order of local variables from longest to shortest line
- fix ave_probe() which calls register_netdev() at the end of initialization
- dt-bindings: remove phy node descriptions in mdio node

Changes since v5:
- replace license boilerplate with SPDX Identifier
- remove inline directives and an unused function

Changes since v4:
- fix larger integer warning on AVE_PFMBYTE_MASK0

Changes since v3:
- remove checking dma address and use dma_set_mask() to restirct address
- replace ave_mdio_busywait() with read_poll_timeout()
- replace functions to access to registers with readl/writel() directly
- replace a function to access to macaddr with ave_hw_write_macaddr()
- change return value of ave_dma_map() to error value
- move mdiobus_unregister() from ave_remove() to ave_uninit()
- eliminate else block at the end of ave_dma_map()
- add mask definitions for packet filter
- sort bitmap definitions in descending order
- add error check to some functions
- rename and sort functions to clear sub-categories
- fix error value consistency
- remove unneeded initializers
- change type of constant arrays

Changes since v2:
- replace clk_get() with devm_clk_get()
- replace reset_control_get() with devm_reset_control_get_optional_shared()
- add error return when the error occurs on the above *_get functions
- sort soc data and compatible strings
- remove clearly obvious comments
- modify dt-bindings document consistent with these modifications

Changes since v1:
- add/remove devicetree properties and sub-node
  - remove "internal-phy-interrupt" and "desc-bits" property
  - add SoC data structures based on compatible strings
  - add node operation to apply "mdio" sub-node
- add support for features
  - add support for {get,set}_pauseparam and pause frame operations
  - add support for ndo_get_stats64 instead of ndo_get_stats
- replace with desiable functions
  - replace check for valid phy_mode with phy_interface{_mode}_is_rgmii()
  - replace phy attach message with phy_attached_info()
  - replace 32bit operation with {upper,lower}_32_bits() on ave_wdesc_addr()
  - replace nway_reset and get_link with generic functions
- move operations to proper functions
  - move phy_start_aneg() to ndo_open,
    and remove unnecessary PHY interrupt operations
    See http://www.spinics.net/lists/netdev/msg454590.html
  - move irq initialization and descriptor memory allocation to ndo_open
  - move initialization of reset and clock and mdiobus to ndo_init
- fix skbuffer operations
  - fix skb alignment operations and add Rx buffer adjustment for descriptor
    See http://www.spinics.net/lists/netdev/msg456014.html


  - add error returns when dma_map_single() failed
- clean up code structures
  - clean up wait-loop and wake-queue conditions
  - add ave_wdesc_addr() and offset definitions
  - add ave_macaddr_init() to clean up mac-address operation
  - fix checking whether Tx entry is not enough
  - fix supported features of phydev
  - add necessary free/disable operations
  - add phydev check on ave_{get,set}_wol()
  - remove netif_carrier functions, phydev initializer, and Tx budget check
- change obsolate codes
  - replace ndev->{base_addr,irq} with the members of ave_private
- rename goto labels and mask definitions, and remove unused codes
====================

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

The UniPhier platform from Socionext provides the AVE ethernet
controller that includes MAC and MDIO bus supporting RGMII/RMII
modes. The controller is named AVE.

Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Signed-off-by: Jassi Brar <jaswinder.singh@linaro.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

DT bindings for the AVE ethernet controller found on Socionext's
UniPhier platforms.

Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Signed-off-by: Jassi Brar <jaswinder.singh@linaro.org>
Acked-by: Rob Herring <robh@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for new Backplane XLAUI port type.

Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix incorrect VNI display in mps_tcam

Signed-off-by: Santosh Rastapur <santosh@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Ethernet representors have a need to store data which is applicable
only for them. Create a priv void pointer in struct mlx5_eswitch_rep
and move mlx5e to store the relevant data there. As part of this change
we also initialize rep_if in mlx5e_rep_register_vf_vports() as otherwise the
E-Switch code will copy a priv value which is garbage.

We also rename mlx5_eswitch_get_uplink_netdev() to
mlx5_eswitch_get_uplink_priv() and make it return void *.
This way E-Switch code doesn't need to deal with net devices and
we leave the task of getting it to mlx5e.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

In order for representors to send packets directly to VFs we use an
E-Switch function which insert special rules into the HW. For symmetry
create an E-Switch function that deletes these rules as well.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

In our pursuit to cleanup e-switch sub-module from mlx5e specific code,
we move the functions that insert/remove the flow steering rules that
allow mlx5e representors to send packets directly to VFs into the EN
driver code.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

In the load() callback for loading representors we don't really need
struct mlx5_eswitch but struct mlx5_core_dev, pass it directly.

In the unload() callback for unloading representors we don't need the
struct mlx5_eswitch argument, remove it.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Refactor the load/unload stages for better code reuse.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Refactor the init stage of vport representors registration.
vport number and hw id can be assigned by the E-Switch driver and not by
the netdevice driver. While here, make the error path of mlx5_eswitch_init()
a reverse order of the good path, also use kcalloc to allocate an array
instead of kzalloc.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Fixes: 482d2e9c

 ("net: hns3: add support to query tqps number")
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Daniel Borkmann says:

====================
pull-request: bpf-next 2017-12-28

The following pull-request contains BPF updates for your *net-next* tree.

The main changes are:

1) Fix incorrect state pruning related to recognition of zero initialized
   stack slots, where stacksafe exploration would mistakenly return a
   positive pruning verdict too early ignoring other slots, from Gianluca.

2) Various BPF to BPF calls related follow-up fixes. Fix an off-by-one
   in maximum call depth check, and rework maximum stack depth tracking
   logic to fix a bypass of the total stack size check reported by Jann.
   Also fix a bug in arm64 JIT where prog->jited_len was uninitialized.
   Addition of various test cases to BPF selftests, from Alexei.

3) Addition of a BPF selftest to test_verifier that is related to BPF to
   BPF calls which demonstrates a late caller stack size increase and
   thus out of bounds access. Fixed above in 2). Test case from Jann.

4) Addition of correlating BPF helper calls, BPF to BPF calls as well
   as BPF maps to bpftool xlated dump in order to allow for better
   BPF program introspection and debugging, from Daniel.

5) Fixing several bugs in BPF to BPF calls kallsyms handling in order
   to get it actually to work for subprogs, from Daniel.

6) Extending sparc64 JIT support for BPF to BPF calls and fix a couple
   of build errors for libbpf on sparc64, from David.

7) Allow narrower context access for BPF dev cgroup typed programs in
   order to adapt to LLVM code generation. Also adjust memlock rlimit
   in the test_dev_cgroup BPF selftest, from Yonghong.

8) Add netdevsim Kconfig entry to BPF selftests since test_offload.py
   relies on netdevsim device being available, from Jakub.

9) Reduce scope of xdp_do_generic_redirect_map() to being static,
   from Xiongwei.

10) Minor cleanups and spelling fixes in BPF verifier, from Colin.
====================

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

After TC offloads were converted to callbacks we have no choice
but keep track of the offloaded filter in the driver.

Since this change came a little late in the release cycle
there were a number of conflicts and allocation of vNIC priv
structure seems to have slipped away in linux-next.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Daniel Borkmann says:

====================
pull-request: bpf 2017-12-28

The following pull-request contains BPF updates for your *net* tree.

The main changes are:

1) Two small fixes for bpftool. Fix otherwise broken output if any of
   the system calls failed when listing maps in json format and instead
   of bailing out, skip maps or progs that disappeared between fetching
   next id and getting an fd for that id, both from Jakub.

2) Small fix in BPF selftests to respect LLC passed from command line
   when testing for -mcpu=probe presence, from Quentin.
====================

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

ibmr.device is being set only after ib_alloc_mr() is
(successfully) complete. Therefore, in case mlx5_core_create_mkey()
return with error, the error flow calls mlx5_free_priv_descs()
which uses ibmr.device (which doesn't exist yet), causing
a NULL dereference oops.

To fix this, the IB device should be set in the mr struct earlier
stage (e.g. prior to calling mlx5_core_create_mkey()).

Fixes: 8a187ee5

 ("IB/mlx5: Support the new memory registration API")
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Nitzan Carmi <nitzanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>