While the nmi_enter() users did
trace_hardirqs_{off_prepare,on_finish}() there was no matching
lockdep_hardirqs_*() calls to complete the picture.
Introduce idtentry_{enter,exit}_nmi() to enable proper IRQ state
tracking across the NMIs.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200623083721.216740948@infradead.org
In order to use <asm/percpu.h> in lockdep.h, we need to make sure
asm/percpu.h does not itself depend on lockdep.
The below seems to make that so and builds powerpc64-defconfig +
PROVE_LOCKING.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
https://lkml.kernel.org/r/20200623083721.336906073@infradead.org
There is no reason not to always, accurately, track IRQ state.
This change also makes IRQ state tracking ignore lockdep_off().
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200623083721.155449112@infradead.org
In order to break a header dependency between lockdep and task_struct,
I need per-cpu stuff from lockdep.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Link: https://lkml.kernel.org/r/20200623083721.277992771@infradead.org
The new IRQ state tracking code does not honor lockdep_off(), and as
such we should again permit tracing by using non-raw functions in
core.c. Update the lockdep_off() comment in report.c, to reflect the
fact there is still a potential risk of deadlock due to using printk()
from scheduler code.
Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200624113246.GA170324@elver.google.com
Perf implements a GPU delay (noa_wait) by looping until the CS timestamp
has passed a certain point. This use MI_MATH and the general purpose
registers of the user's context, and since it is clobbering the user
state it must carefully save and restore the user's data around the
noa_wait. We can verify this by loading some values in the GPR that we
know will be clobbered by the noa_wait, and then inspecting the GPR after
the noa_wait completes and confirming that they have been restored.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200709224504.11345-2-chris@chris-wilson.co.uk
MI_STORE_REGISTER_MEM and MI_LOAD_REGISTER_MEM need to know which
translation to use when saving restoring the engine general purpose
registers to and from the GT scratch. Since GT scratch is mapped to
ggtt, we need to set an additional bit in the command to use GTT.
Fixes: daed3e4439 ("drm/i915/perf: implement active wait for noa configurations")
Suggested-by: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20200709224504.11345-1-chris@chris-wilson.co.uk
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In the current kvm version, 'kvm_run' has been included in the 'kvm_vcpu'
structure. For historical reasons, many kvm-related function parameters
retain the 'kvm_run' and 'kvm_vcpu' parameters at the same time. This
patch does a unified cleanup of these remaining redundant parameters.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Reviewed-by: Huacai Chen <chenhc@lemote.com>
Message-Id: <20200623131418.31473-5-tianjia.zhang@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In the current kvm version, 'kvm_run' has been included in the 'kvm_vcpu'
structure. For historical reasons, many kvm-related function parameters
retain the 'kvm_run' and 'kvm_vcpu' parameters at the same time. This
patch does a unified cleanup of these remaining redundant parameters.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200623131418.31473-3-tianjia.zhang@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In the current kvm version, 'kvm_run' has been included in the 'kvm_vcpu'
structure. For historical reasons, many kvm-related function parameters
retain the 'kvm_run' and 'kvm_vcpu' parameters at the same time. This
patch does a unified cleanup of these remaining redundant parameters.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200623131418.31473-2-tianjia.zhang@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This device hierarchy is needlessly complex.
Remove the support-card node level, and move the ethernet and serial
nodes right under the system-bus node.
This also fixes the following warning from 'make ARCH=arm dtbs_check':
support-card@1,1f00000: $nodename:0: 'support-card@1,1f00000' does not match '^(bus|soc|axi|ahb|apb)(@[0-9a-f]+)?$'
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
OP-TEE based fTPM Trusted Application depends on tee-supplicant to
provide NV RAM implementation based on RPMB secure storage. So this
dependency can be resolved via TEE bus where we only invoke fTPM
driver probe once fTPM device is registered on the bus which is only
true after the tee-supplicant is up and running. Additionally, TEE bus
provides auto device enumeration.
Signed-off-by: Maxim Uvarov <maxim.uvarov@linaro.org>
Suggested-by: Sumit Garg <sumit.garg@linaro.org>
Suggested-by: Arnd Bergmann <arnd@linaro.org>
Reviewed-by: Sumit Garg <sumit.garg@linaro.org>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>
Some drivers (like ftpm) can operate only after tee-supplicant
runs because of tee-supplicant provides things like storage
services (rpmb, shm). This patch splits probe of non tee-supplicant
dependable drivers to the early stage, and after tee-supplicant run
probe other drivers.
Signed-off-by: Maxim Uvarov <maxim.uvarov@linaro.org>
Suggested-by: Sumit Garg <sumit.garg@linaro.org>
Suggested-by: Arnd Bergmann <arnd@linaro.org>
Reviewed-by: Sumit Garg <sumit.garg@linaro.org>
Acked-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>
With the evolving use-cases for TEE bus, now it's required to support
multi-stage enumeration process. But using a simple index doesn't
suffice this requirement and instead leads to duplicate sysfs entries.
So instead switch to use more informative device UUID for sysfs entry
like:
/sys/bus/tee/devices/optee-ta-<uuid>
Signed-off-by: Maxim Uvarov <maxim.uvarov@linaro.org>
Reviewed-by: Sumit Garg <sumit.garg@linaro.org>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Sumit Garg <sumit.garg@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>
This file was created long ago and information inside is
obviously outdated.
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Message-Id: <20200710063047.154611-4-jiaxun.yang@flygoat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
After tons of fixes to get Trap-and-Emulate build on Loongson64,
I've got panic on host machine when trying to run a VM.
I found that it can never work on 64bit systems. Revewing the
code, it looks like R6 can't supportrd by TE as well.
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Message-Id: <20200710063047.154611-3-jiaxun.yang@flygoat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
clang static analysis flags this error
c67x00-sched.c:489:55: warning: Use of memory after it is freed [unix.Malloc]
usb_hcd_giveback_urb(c67x00_hcd_to_hcd(c67x00), urb, urbp->status);
^~~~~~~~~~~~
Problem happens in this block of code
c67x00_release_urb(c67x00, urb);
usb_hcd_unlink_urb_from_ep(c67x00_hcd_to_hcd(c67x00), urb);
spin_unlock(&c67x00->lock);
usb_hcd_giveback_urb(c67x00_hcd_to_hcd(c67x00), urb, urbp->status);
In the call to c67x00_release_urb has this freeing of urbp
urbp = urb->hcpriv;
urb->hcpriv = NULL;
list_del(&urbp->hep_node);
kfree(urbp);
And so urbp is freed before usb_hcd_giveback_urb uses it as its 3rd
parameter.
Since all is required is the status, pass the status directly as is
done in c64x00_urb_dequeue
Fixes: e9b29ffc51 ("USB: add Cypress c67x00 OTG controller HCD driver")
Signed-off-by: Tom Rix <trix@redhat.com>
Cc: stable <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200708131243.24336-1-trix@redhat.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The sparse tool complains as follows:
drivers/usb/gadget/udc/lpc32xx_udc.c:2989:25: warning:
symbol 'lpc32xx_usbddata' was not declared. Should it be static?
This variable is not used outside of lpc32xx_udc.c, so this commit
marks it static.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Link: https://lore.kernel.org/r/20200707105350.7064-1-weiyongjun1@huawei.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Xin Long says:
==================
Now ipip and ipv6 tunnels processing is supported by xfrm4/6_tunnel,
but not in vti and xfrmi. This feature is needed by processing those
uncompressed small fragments and packets when using comp protocol.
It means vti and xfrmi won't be able to accept small fragments or
packets when using comp protocol, which is not expected.
xfrm4/6_tunnel eventually calls xfrm_input() to process ipip and ipv6
tunnels with an ipip/ipv6-proto state (a child state of comp-proto
state), and vti and xfrmi should do the same.
The extra things for vti to do is:
- vti_input() should be called before xfrm_input() to set
XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4/6 = tunnel. [A]
- vti_rcv_cb() should be called after xfrm_input() to update
the skb->dev. [B]
And the extra things for xfrmi to do is:
- The ipip/ipv6-proto state should be assigned if_id from its
parent's state. [C]
- xfrmi_rcv_cb() should be called after xfrm_input() to update
the skb->dev. [D]
Patch 4-7 does the things in [A].
To implement [B] and [D], patch 1-3 is to build a callback function
for xfrm4/6_tunnel, which can be called after xfrm_input(), similar
to xfrm4/6_protocol's .cb_handler. vti and xfrmi only needs to give
their own callback function in patch 4-7 and 9-10, which already
exists: vti_rcv_cb() and xfrmi_rcv_cb().
Patch 8 is to do the thing in [C] by assigning child tunnel's if_id
from its parent tunnel.
With the whole patch series, the segments or packets with any size
can work with ipsec comp proto on vti and xfrmi.
v1->v2:
- See Patch 2-3.
v2->v3:
- See Patch 2-3, 4, 6, 9-10.
==================
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
The constant arrays in gdb_xml.h are only used in arch/riscv/kernel/kgdb.c,
but other c files may include the gdb_xml.h indirectly via including the
kgdb.h. Hence, It will cause many unused-const-variable warnings. This
patch makes the kgdb.h not to include the gdb_xml.h to solve this problem.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Currently, only riscv kgdb.c uses the kgdb_has_hit_break() to identify
the kgdb breakpoint. It causes other architectures will encounter the "no
previous prototype" warnings if the compile option has W=1. Moving the
declaration of extern kgdb_has_hit_break() from risc-v kgdb.h to generic
kgdb.h to avoid generating these warnings.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Acked-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Some functions are only used in the kgdb.c file. Add static properities
to these functions to avoid "no previous prototype" compile warnings
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Due to lack of hardware breakpoint support, the kernel option
CONFIG_STRICT_KERNEL_RWX should be disabled when using KGDB. However,
CONFIG_STRICT_KERNEL_RWX is always enabled now. Therefore, select
ARCH_OPTIONAL_KERNEL_RWX_DEFAULT to enable CONFIG_STRICT_KERNEL_RWX
by default, and then select ARCH_OPTIONAL_KERNEL_RWX to enable the
Kconfig prompt of CONFIG_STRICT_KERNEL_RWX so that users can turn it off.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
The XML packet could be supported by required architecture if the
architecture defines CONFIG_HAVE_ARCH_KGDB_QXFER_PKT and implement its own
kgdb_arch_handle_qxfer_pkt(). Except for the kgdb_arch_handle_qxfer_pkt(),
the architecture also needs to record the feature supported by gdb stub
into the kgdb_arch_gdb_stub_feature, and these features will be reported
to host gdb when gdb stub receives the qSupported packet.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Acked-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Before this commit, on ft flush, ft entries were not removed
from the ct_tuple hashtables. Fix it.
Fixes: ac991b48d4 ("net/mlx5e: CT: Offload established flows")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
"flow" parameter is not used in __mlx5_tc_ct_flow_offload_clear(),
remove it.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Instead of having to deal with converting between int and ERR_PTR for
return values in mlx5_tc_ct_flow_offload(), make the internal helper
functions return a ptr to mlx5_flow_handle instead of passing it as
output param, this will also avoid gcc confusion and false alarms,
thus we remove the redundant ERR_PTR rule initialization.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Suggested-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Reg_c1 is 32 bits wide. Originally, 24 bit were allocated for the tuple_id,
6 bits for tunnel mapping and 2 bits for tunnel options mappings.
Restoring the ct state from zone lookup instead of tuple id requires
reg_c1 to store 8 bits mapping the ct zone, leaving 24 bits for tunnel
mappings.
Expand tunnel and tunnel options register mappings to 12 bit each.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Use a single byte mapping for zone restore register (zone matching
remains 16 bit).
This makes room for using the freed 8 bits on register C1 for
mapping more tunnels and tunnel options.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
After removing the tupleid register which changed per tuple,
tuple modify headers set the ct_state, zone, mark, and label registers.
For non-natted tuples going through the same tc rules path, their values
will be the same, and all their modify headers will be the same.
Re-use tuple modify header when possible, by adding each new modify
header to an hahstable, and looking up identical ones before creating
a new one.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Refactor sharing of mod headers to new file and while there,
remove spin lock and flows list, as this is only used for warn on.
Use the generic API in the next patch to re-use tuple modify headers
for identical modify actions,
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Remove tupleid, and replace it with zone_restore, which is the zone an
established tuple sets after match. On miss, Use this zone + tuple
taken from the skb, to lookup the ct entry and restore it.
This improves flow insertion rate by avoiding the allocation of a header
rewrite context.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Next patches will remove the tupleid registers that is used
to restore the ct state on miss, and instead use the tuple on
the missed packet to lookup which state to restore.
Disable tuple rewrites after connection tracking.
For tuple rewrites, inject a ct_state=-trk match so it won't
change the tuple for established flows (+trk) that passed connection
tracking, and instead miss to software.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
The next patch will pass the mlx5e_priv struct to the
modify_header_match_supported method. Use this opportunity to refactor
the existing pr_info call to a netdev_info call.
Signed-off-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
With ct clear we don't jump to the ct tables, so header rewrite
of 5-tuple can be done in place (and not moved to after the CT action).
Check for ct clear action, and if so, allow 5-tuple header
rewrite.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Save original tuple and natted tuple in two new hashtables.
This is a pre-step for restoring ct state after hw miss by performing a
5-tuple lookup on the hash tables.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
When eswitch is unsupported, currently -EPERM error code is returned
instead of -EOPNOTSUPP.
Due to this VF device's devlink virtual port is not enumerated because
port_function_get() callback returned -EPERM instead of -EOPNOTSUPP.
Hence, return the error code -EOPNOTSUPP when eswitch is unsupported.
Fixes: bd93975353 ("net/mlx5: E-switch, Introduce and use eswitch support check helper")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
On ILP32, 64-bit result was shifted by value calculated for 32-bit long type
and returned value was much outside hashmap capacity.
As advised by Andrii Nakryiko, this patch uses different hashing variant for
architectures with size_t shorter than long long.
Fixes: e3b9242240 ("libbpf: add resizable non-thread safe internal hashmap")
Signed-off-by: Jakub Bogusz <qboosh@pld-linux.org>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200709225723.1069937-1-andriin@fb.com
CT entries are deleted via a workqueue from netfilter. If removing the
module before that, the rules are cleaned by the driver itself, but the
memory entries for them are not freed. Fix that.
Fixes: ac991b48d4 ("net/mlx5e: CT: Offload established flows")
Signed-off-by: Eli Britstein <elibr@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Device unit for port buffers size, xoff_threshold and xon_threshold is
cells. Fix a bug in driver where cell unit size was hard-coded to
128 bytes. This hard-coded value is buggy, as it is wrong for some hardware
versions.
Driver to read cell size from SBCAM register and translate bytes to cell
units accordingly.
In order to fix the bug, this patch exposes SBCAM (Shared buffer
capabilities mask) layout and defines.
If SBCAM.cap_cell_size is valid, use it for all bytes to cells
calculations. If not valid, fallback to 128.
Cell size do not change on the fly per device. Instead of issuing SBCAM
access reg command every time such translation is needed, cache it in
mlx5e_dcbx as part of mlx5e_dcbnl_initialize(). Pass dcbx.port_buff_cell_sz
as a param to every function that needs bytes to cells translation.
While fixing the bug, move MLX5E_BUFFER_CELL_SHIFT macro to
en_dcbnl.c, as it is only used by that file.
Fixes: 0696d60853 ("net/mlx5e: Receive buffer configuration")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: Huy Nguyen <huyn@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Some released FW versions mistakenly don't set the capability that 50G
per lane link-modes are supported for VFs (ptys_extended_ethernet
capability bit). When the capability is unset, read
PTYS.ext_eth_proto_capability (always reliable).
If PTYS.ext_eth_proto_capability is valid (has a non-zero value)
conclude that the HCA supports 50G per lane. Otherwise, conclude that
the HCA doesn't support 50G per lane.
Fixes: a08b4ed137 ("net/mlx5: Add support to ext_* fields introduced in Port Type and Speed register")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
When detaching netdev, remove vxlan port configuration using
udp_tunnel_drop_rx_info. During function reload, configuration will be
restored using udp_tunnel_get_rx_info. This ensures sync between
firmware and driver. Use udp_tunnel_get_rx_info even if its physical
interface is down.
Fixes: 4383cfcc65 ("net/mlx5: Add devlink reload")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Peter Zijlstra