This series comes to allow user space applications to monitor real time
traffic activity and events of the verbs objects it manages, e.g.:
ibv_qp, ibv_wq, ibv_flow.
This API enables generic counters creation and define mapping
to association with a verbs object, current mlx5 driver using
this API for flow counters.
With this API, an application can monitor the entire life cycle of
object activity, defined here as a static counters attachment.
This API also allows dynamic counters monitoring of measurement points
for a partial period in the verbs object life cycle.
In addition it presents the implementation of the generic counters interface.
This will be achieved by extending flow creation by adding a new flow count
specification type which allows the user to associate a previously created
flow counters using the generic verbs counters interface to the created flow,
once associated the user could read statistics by using the read function of
the generic counters interface.
The API includes:
1. create and destroyed API of a new counters objects
2. read the counters values from HW
Note:
Attaching API to allow application to define the measurement points per objects
is a user space only API and this data is passed to kernel when the counted
object (e.g. flow) is created with the counters object.
-----BEGIN PGP SIGNATURE-----
iHQEABYIAB0WIQT1m3YD37UfMCUQBNwp8NhrnBAZsQUCWxIiqQAKCRAp8NhrnBAZ
sWJRAPYl06nEfQjRlW//ZE/pO2oKXbfEevg7nnbpe80ERlxLAQDA2LHAcU7ma/NC
hS5yxIq1gLSA27N+5qAoFVK8vJ5ZCg==
=EiAV
-----END PGP SIGNATURE-----
Merge tag 'verbs_flow_counters' of git://git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git into for-next
Pull verbs counters series from Leon Romanovsky:
====================
Verbs flow counters support
This series comes to allow user space applications to monitor real time
traffic activity and events of the verbs objects it manages, e.g.: ibv_qp,
ibv_wq, ibv_flow.
The API enables generic counters creation and define mapping to
association with a verbs object, the current mlx5 driver is using this API
for flow counters.
With this API, an application can monitor the entire life cycle of object
activity, defined here as a static counters attachment. This API also
allows dynamic counters monitoring of measurement points for a partial
period in the verbs object life cycle.
In addition it presents the implementation of the generic counters
interface.
This will be achieved by extending flow creation by adding a new flow
count specification type which allows the user to associate a previously
created flow counters using the generic verbs counters interface to the
created flow, once associated the user could read statistics by using the
read function of the generic counters interface.
The API includes:
1. create and destroyed API of a new counters objects
2. read the counters values from HW
Note:
Attaching API to allow application to define the measurement points per
objects is a user space only API and this data is passed to kernel when
the counted object (e.g. flow) is created with the counters object.
===================
* tag 'verbs_flow_counters':
IB/mlx5: Add counters read support
IB/mlx5: Add flow counters read support
IB/mlx5: Add flow counters binding support
IB/mlx5: Add counters create and destroy support
IB/uverbs: Add support for flow counters
IB/core: Add support for flow counters
IB/core: Support passing uhw for create_flow
IB/uverbs: Add read counters support
IB/core: Introduce counters read verb
IB/uverbs: Add create/destroy counters support
IB/core: Introduce counters object and its create/destroy
IB/uverbs: Add an ib_uobject getter to ioctl() infrastructure
net/mlx5: Export flow counter related API
net/mlx5: Use flow counter pointer as input to the query function
This patch implements the uverbs counters read API, it will use the
specific read counters function to the given type to accomplish its
task.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Associates a counters with a flow when IB_FLOW_SPEC_ACTION_COUNT is part
of the flow specifications.
The counters user space placements of location and description (index,
description) pairs are passed as private data of the counters flow
specification.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This patch implements the device counters create and destroy APIs and
introducing some internal management structures.
Downstream patches in this series will add the functionality to support
flow counters binding and reading.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The struct ib_uverbs_flow_spec_action_count associates a counters object
with the flow.
Post this association the flow counters can be read via the counters
object.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
A counters object could be attached to flow on creation by providing the
counter specification action.
General counters description which count packets and bytes are introduced,
downstream patches from this series will use them as part of flow counters
binding.
In addition, increase number of flow specifications supported layers to 10
upon adding count specification and for the previously added drop
specification.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This is required when user-space drivers need to pass extra information
regarding how to handle this flow steering specification.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This patch exposes the read counters verb to user space applications. By
that verb the user can read the hardware counters which are associated
with the counters object.
The application needs to provide a sufficient memory to hold the
statistics.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The user supplies counters instance and a reference to an output array of
uint64_t. The driver reads the hardware counters values and writes them
to the output index location in the user supplied array. All counters
values are represented as uint64_t types.
To be able to successfully read the data the counters must be first bound
to an IB object.
Downstream patches will present binding method for flow counters.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
User space application which uses counters functionality, is expected to
allocate/release the counters resources by calling create/destroy verbs
and in turn get a unique handle that can be used to attach the counters to
its counted type.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
A verbs application may need to get statistics and info on various aspects
of a verb object (e.g. Flow, QP, ...), in general case the application
will state which object's counters its interested in (we refer to this
action as attach), bind this new counters object to the appropriate verb
object and on later stage read their values using the counters object.
This series introduces a general API for counters object that may
accumulate any ib object counters type, bound and read on demand.
Counters instance is allocated on an IB context and belongs to that
context. Upon successful creation the counters can be bound to a verbs
object so that hardware counter instances can be created and read.
Downstream patches in this series will introduce the attach, bind and the
read functionality.
Counters instance can be de-allocated, upon successful destruction the
related hardware resources are released.
Prior to destroy call the user must first make sure that the counters is
not being used by any IB object, e.g. not attached to any of its counted
type otherwise an EBUSY error is invoked.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Previously, the user had to dig inside the attribute to get the uobject.
Add a helper function that correctly extract it (and do the required
checks) for him/her.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Exports counters API to be used in both IB and EN.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This allows to un-expose the details of struct mlx5_fc and keep it
internal to the core driver.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
All callers of zap_vma_ptes() are not interested in the return value of
that function, so let's simplify its interface and drop the return
value.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is no need to check return value of zap_vma_ptes()
because there is nothing to do with this knowledge.
Signed-off-by: Doug Ledford <dledford@redhat.com>
The failure reported by zap_vma_ptes() means that wrong VMA pages
were supplied, however it is impossible for this type of address.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is no need to check return value of zap_vma_ptes()
because there is nothing to do with this knowledge.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is no need to check existence of mad_queue, because we already did
pointer dereference before call to dequeue_mad().
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is no need to bring down the whole machine, just because unknown
event was received. It is better to ignore it silently.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In case CM work list is empty, the work pointer will be NULL,
so instead of kernel crash it is better to abort processing
of works.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
cxgb3 driver properly handles errors returned by IDR, so there is no
need to have special case (kernel crash) just because IDR is full.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is no need to crash the machine if unknown work request was
received in SQP MAD.
Cc: <stable@vger.kernel.org> # 3.6
Fixes: 37bfc7c1e8 ("IB/mlx4: SR-IOV multiplex and demultiplex MADs")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Any steering QP is supposed be above steering_qp_base,
see function mlx4_ib_steer_qp_alloc() for it, however in case
of misalignment between SW and FW, this qp_base can be wrong.
Use WARN() to catch such situation without killing the machine.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Trivial fix to spelling mistake in DP_ERR error message
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The iSER driver reduces max_sectors. For example, if you load the
ib_iser module with max_sectors=1024, you will see that
/sys/class/block/<bdev>/queue/max_hw_sectors_kb is 508. It is an
incorrect value. The expected value is (max_sectors * sector_size) /
1024 = 512.
Reducing of max_sectors can cause performance degradation due to
unnecessary splitting of IO requests.
The number of pages per MR has been fixed here, so there is no longer
any need to reduce max_sectors.
Fixes: 9c674815d3 ("IB/iser: Fix max_sectors calculation")
Signed-off-by: Sergey Gorenko <sergeygo@mellanox.com>
Reviewed-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Sagi Grimberg <sagi@grimberg.me>
Acked-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch hoisted the common process of disassociate_ucontext
callback function into ib core code, and these code are common
to ervery ib_device driver.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch added reset process for RoCE in hip08.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Leon Romanovsky says:
====================
Introduce new internal to mlx5 CQE format - mini-CQE. It is a CQE in
compressed form that holds data needed to extra a single full CQE.
It is a stride index, byte count and packet checksum.
====================
* mini_cqe:
IB/mlx5: Introduce a new mini-CQE format
IB/mlx5: Refactor CQE compression response
net/mlx5: Exposing a new mini-CQE format
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This once might have made sense when cache.c was in a different module
from device.c, but today it just obfuscation. Get rid of the wrappers
and call roge_gid_mgmt_init()/cleanup() directly.
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
The new mini-CQE format includes the stride index, byte count and
packet checksum.
Stride index is needed for striding WQ feature.
This patch exposes this capability and enables its setting
via mlx5 UHW data as part of query device and cq creation.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Refactor CQE compression response to be fully set only
when it`s really supported. There is no change from user
perspective because anyway resp.cqe_comp_caps.max_num was
set to zero.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>W
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The new mini-CQE format includes byte-count, checksum
and stride index.
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Update mlx4 to support user MR creation against read-only memory, previously
it required the memory to be writable.
Based on rdma for-rc due to dependencies.
* mr_fix: (2 commits)
IB/mlx4: Mark user MR as writable if actual virtual memory is writable
IB/core: Make testing MR flags for writability a static inline function
To allow rereg_user_mr to modify the MR from read-only to writable without
using get_user_pages again, we needed to define the initial MR as writable.
However, this was originally done unconditionally, without taking into
account the writability of the underlying virtual memory.
As a result, any attempt to register a read-only MR over read-only
virtual memory failed.
To fix this, do not add the writable flag bit when the user virtual memory
is not writable (e.g. const memory).
However, when the underlying memory is NOT writable (and we therefore
do not define the initial MR as writable), the IB core adds a
"force writable" flag to its user-pages request. If this succeeds,
the reg_user_mr caller gets a writable copy of the original pages.
If the user-space caller then does a rereg_user_mr operation to enable
writability, this will succeed. This should not be allowed, since
the original virtual memory was not writable.
Cc: <stable@vger.kernel.org>
Fixes: 9376932d0c ("IB/mlx4_ib: Add support for user MR re-registration")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Make the MR writability flags check, which is performed in umem.c,
a static inline function in file ib_verbs.h
This allows the function to be used by low-level infiniband drivers.
Cc: <stable@vger.kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
The function rxe_remove_all is only used in this modules.
There is no other modules that call this function. So it
is not necessary to export it.
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This patch increases checking CMQ status timeout value and
uses the same value with NIC driver to avoid deficiency of
time.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This patch modified uar allocation algorithm in hns_roce_uar_alloc
function to avoid bitmap exhaust.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
mlx5 core dirver updates for both net-next and rdma-next branches.
From Christophe JAILLET, first three patche to use kvfree where needed.
From: Or Gerlitz <ogerlitz@mellanox.com>
Next six patches from Roi and Co adds support for merged
sriov e-switch which comes to serve cases where both PFs, VFs set
on them and both uplinks are to be used in single v-switch SW model.
When merged e-switch is supported, the per-port e-switch is logically
merged into one e-switch that spans both physical ports and all the VFs.
This model allows to offload TC eswitch rules between VFs belonging
to different PFs (and hence have different eswitch affinity), it also
sets the some of the foundations needed for uplink LAG support.
-----BEGIN PGP SIGNATURE-----
iQEcBAABAgAGBQJa/fLEAAoJEEg/ir3gV/o+7jUH/3n5/Uw1LLt3TfeKArx6i0F1
3G4U5B0ha03qiDqXprwhyQ3I6lgYmRBmjcxnqmvcqOAqO4/hSsjtTR+A/mgbEDhJ
YtdekFNEX+72h/N2GIpZwChIWSE3EcMPaLYnV8TwLUgh9YSust2sCLSBbJCjxOKc
j78M8ept/bXZwTm/iJhEjtmqw0xl91rl011chCAua0iEpH3wxteDARmKABFHMQxl
I3N/x/e/astgcSCNgpO4uDf9zEIRkNdzcHPzSMJ6C2Oo5W9XiZEekfw7WKj9nXfa
G+eGckkAyCOQ/r2lZ9nA0ZUvQ2X6JISvxgohuaCNwTgsz3acTxbLnQK4YWHzQCQ=
=iHi6
-----END PGP SIGNATURE-----
Merge tag 'mlx5-updates-2018-05-17' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux into for-next
mlx5-updates-2018-05-17
mlx5 core dirver updates for both net-next and rdma-next branches.
From Christophe JAILLET, first three patche to use kvfree where needed.
From: Or Gerlitz <ogerlitz@mellanox.com>
Next six patches from Roi and Co adds support for merged
sriov e-switch which comes to serve cases where both PFs, VFs set
on them and both uplinks are to be used in single v-switch SW model.
When merged e-switch is supported, the per-port e-switch is logically
merged into one e-switch that spans both physical ports and all the VFs.
This model allows to offload TC eswitch rules between VFs belonging
to different PFs (and hence have different eswitch affinity), it also
sets the some of the foundations needed for uplink LAG support.
* tag 'mlx5-updates-2018-05-17' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
net/mlx5e: Explicitly set source e-switch in offloaded TC rules
net/mlx5: Add source e-switch owner
net/mlx5e: Explicitly set destination e-switch in FDB rules
net/mlx5: Add destination e-switch owner
net/mlx5: Properly handle a vport destination when setting FTE
net/mlx5: Add merged e-switch cap
IB/mlx5: Use 'kvfree()' for memory allocated by 'kvzalloc()'
net/mlx5: Eswitch, Use 'kvfree()' for memory allocated by 'kvzalloc()'
net/mlx5: Vport, Use 'kvfree()' for memory allocated by 'kvzalloc()'
There are several places a gid table is accessed.
Have a helper tiny function rdma_gid_table() to avoid code
duplication at such places.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Instead of open coding memcmp() to check whether a given GID is zero or
not, use a helper function to do so, and replace instances of
memcpy(z,&zgid) with memset.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
On fatal error the driver simulates CQE's for ULPs that rely on
completion of all their posted work-request.
For the GSI traffic, the mlx5 has its own mechanism that sends the
completions via software CQE's directly to the relevant CQ.
This should be kept in fatal error too, so the driver should simulate
such CQE's with the specified error state in order to complete GSI QP
work requests.
Without the fix the next deadlock might appears:
schedule_timeout+0x274/0x350
wait_for_common+0xec/0x240
mcast_remove_one+0xd0/0x120 [ib_core]
ib_unregister_device+0x12c/0x230 [ib_core]
mlx5_ib_remove+0xc4/0x270 [mlx5_ib]
mlx5_detach_device+0x184/0x1a0 [mlx5_core]
mlx5_unload_one+0x308/0x340 [mlx5_core]
mlx5_pci_err_detected+0x74/0xe0 [mlx5_core]
Cc: <stable@vger.kernel.org> # 4.7
Fixes: 89ea94a7b6 ("IB/mlx5: Reset flow support for IB kernel ULPs")
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
In commit 357d23c811a7 ("Remove the obsolete libibcm library")
in rdma-core [1], we removed obsolete library which used the
/dev/infiniband/ucmX interface.
Following multiple syzkaller reports about non-sanitized
user input in the UCMA module, the short audit reveals the same
issues in UCM module too.
It is better to disable this interface in the kernel,
before syzkaller team invests time and energy to harden
this unused interface.
[1] https://github.com/linux-rdma/rdma-core/pull/279
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Remove duplicate declaration of gid_cache_wq.
Fixes: d41861942 ("IB/core: Add generic function to extract IB speed from netdev")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
