If the RECV CQE is in error, ignore the MSN check. This was causing
recvs that were flushed into the sw cq to be completed with the wrong
status (BAD_MSN instead of FLUSHED).
Cc: stable@vger.kernel.org
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Debugfs file reset_stats is created with S_IRUSR permissions,
but ocrdma_dbgfs_ops_read() doesn't support OCRDMA_RESET_STATS,
whereas ocrdma_dbgfs_ops_write() supports only OCRDMA_RESET_STATS.
The patch fixes misstype with permissions.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Anton Vasilyev <vasilyev@ispras.ru>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When path ah_attr initialization from path record
fails, ib_cm_send_rej() uses av.ah_attr fields to send out reject
message. In such cases initialization of path record software fields
is not needed. Code is simplified for same.
Additionally in current code in cm_req_handler, when ib_get_cached_gid
fails for a given sgid_index of the GID of the GRH of the incoming CM MAD,
error code 12 is sent. This error code refers to primary GID in incoming
CM REQ and not for the GID in in MAD packet.
Therefore code is refactored to send code 5 (unsupported request) for such
error.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Hal Rosenstock <hal@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently ib_init_ah_from_wc initializes address handle attributes and
not the address handle object itself.
To avoid confusion between ah_attr vs ah, ib_init_ah_from_wc is
renamed to ib_init_ah_attr_from_wc to reflect that its initialzes
ah_attr.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Since ib_init_ah_from_path initializes the address handle attribute, it is
renamed to reflect so.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
In case of LAP are used for RoCE, it can lead to a problem of sleeping a
context while spin lock is held in below flow.
cm_lap_handler
->spin_lock
-> <..switch_case..>
-> cm_init_av_for_response
-> ib_init_ah_from_wc
-> rdma_addr_find_l2_eth_by_grh
wait_for_completion()
Therefore ah attribute initialization is done for incoming lap requests
outside of the lock context.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
cm_init_av_by_path depends on ib_init_ah_from_path to initialize ah
attribute and ib_init_ah_from_path() can fail, such error should not
be ignored.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
cm_init_av_for_response depends on ib_init_ah_from_wc() whose return
status is ignored.
ib_init_ah_from_wc() can fail and its return status should be handled as
done in this patch.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently there are no users of ib_find_gid for RoCE transport. It is
only used by IPoIB.
Therefore its simplified to ignore RoCE ports and GID type check which
was previously done for every port.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
rdma_copy_addr copies the ifndex to bound_dev_if.
Therefore avoid copying it again after rdma_copy_addr call is completed.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Since no caller needs vlan, rdma_translate_ip is simplified to avoid
vlan pointer.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
rdma_addr_find_smac_by_sgid() is exported symbol not used by any kernel
module. Therefore its removed.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
rdma_resolve_ip already copies 'addr' to its dev_addr argument.
Remove the duplicate memcpy and since it was the only user, remove the
'addr' member from resolve_cb_context.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
ib_find_gid_by_filter() is used only by ib_core, therefore avoid
exporting it.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
While searching for GID, returned index is not used, so avoid passing
pointer during invocation.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently on every gid entry comparison miss found variable is checked;
which is not needed as those two comparison fail already indicate that
GID is not found yet.
So refactor to avoid such check and copy the GID index when found.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Type cast from void to struct find_gid_index_context is not needed.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Introduce and user helper functions to initialize work for address
resolved and route resolved event that avoid code duplication at few
places.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Avoid setting path record type twice for RoCE.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Current code checks for NULL ndev twice where 2nd check is always
invalid given the fact that during route resolving stage, device address
must be bound to netdevice interface.
This patch simplifies such check.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
As the function name suggests cma_resolve_iboe_route() resolves RoCE
route. However, its default GID type is IB_GID_TYPE_IB and not
IB_GID_TYPE_ROCE, even though both are mapped to the same enum value.
Change default GID type to IB_GID_TYPE_ROCE.
cma_iboe_set_mgid() is updated to reflect the RoCEv2 GID check.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Reviewed-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
In ib_umem structure npages holds original number of sg entries, while
nmap is number of DMA blocks returned by dma_map_sg.
Fixes: c5d76f130b ('IB/core: Add umem function to read data from user-space')
Signed-off-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Add debug prints to the error paths in the connection manager control
flows, to help debug connection management problems.
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The code was using the src size when formatting the dst. They are almost
certainly the same value but it reads wrong.
Fixes: ce117ffac2 ("RDMA/cma: Export AF_IB statistics")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
In case that the PathRecord is not valid (SM changed its network prefix)
ipoib will continue issue PathQuery requests with the same parameters
that are in its database, which are no longer valid anymore.
Now the driver in that case will re-initialize the record from a valid
place (the priv structure keeps the updated values), and a valid request
will be issued.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The ipoib path database is organized around DGIDs from the LLADDR, but the
SA is free to return a different GID when asked for path. This causes a
bug because the SA's modified DGID is copied into the database key, even
though it is no longer the correct lookup key, causing a memory leak and
other malfunctions.
Ensure the database key does not change after the SA query completes.
Demonstration of the bug is as follows
ipoib wants to send to GID fe80:0000:0000:0000:0002:c903:00ef:5ee2, it
creates new record in the DB with that gid as a key, and issues a new
request to the SM.
Now, the SM from some reason returns path-record with other SGID (for
example, 2001:0000:0000:0000:0002:c903:00ef:5ee2 that contains the local
subnet prefix) now ipoib will overwrite the current entry with the new
one, and if new request to the original GID arrives ipoib will not find
it in the DB (was overwritten) and will create new record that in its
turn will also be overwritten by the response from the SM, and so on
till the driver eats all the device memory.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
ib_security_modify_qp and ib_security_pkey_access are core internal
function. So avoid exporting them.
ib_security_pkey_access is used only when secuirty hooks are enabled so
avoid defining it otherwise.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
RoCEv1 does not use the IPv6 stack to resolve the link local DGID since it
uses GID address. It forms the DMAC directly from the DGID.
The code became confused and also tried to use this bypass for RoCEv2
packets, however RoCEv2 always uses a IP address in the GID and must
always use ARP or neighbor discovery to get the DMAC address.
Now that rdma_addr_find_l2_eth_by_grh() supports resolving link local
address to find destination mac address, lets make use of it.
This aligns it to how the rest of the IPv6 stack resolves link local
destination IPv6 address.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When computing a UD reverse path (return AH) from a WC the code was not
doing a route lookup anchored in a specific netdevice. This caused several
bugs, including broken IPv6 link-local address support in RoCEv2. [1]
This fixes the lookup by determining the GID table entry that the HW
matched to the SGID for the WC and then using the netdevice from that
entry to perform the route and ND lookup for the 'DGID' to build a return
AH.
RoCE GID table management ensures that right upper netdevices of the
physical netdevices are added. Therefore init_ah_from_wc doesn't need to
perform such check.
Now that route lookup is done based on the netdevice of the GID entry,
simplify code to not have ifindex and vlan pointers. As part of that,
refactor to have netdevice as input parameter. This is already discussed
at [2].
Finally ib_init_ah_from_wc resolves dmac for unicast GID in similar way as
what ib_resolve_eth_dmac() does. So ib_resolve_eth_dmac is refactored to
split for unicast and non unicast GIDs, so that it can be reused by
ib_init_ah_from_wc.
While we are at refactoring ib_resolve_eth_dmac(), it is further
simplified
(a) to avoid hoplimit as optional parameter, as there is only one
user who always queries hoplimit.
(b) for empty line.
(c) avoided zero initialization of ret.
(d) removed as exported symbol as only ib core uses it.
For IPv6, this is tested using simple rping test as below.
rping -sv -a ::0
rping -c -a fe80::268a:7ff:fe55:4661%ens2f1 -C 1 -v -d
[1] https://www.spinics.net/lists/linux-rdma/msg45690.html
[2] https://www.spinics.net/lists/linux-rdma/msg45710.html
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Reported-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Smatch complains about this code:
drivers/infiniband/hw/mlx4/qp.c:1827 _mlx4_set_path()
error: buffer overflow 'dev->dev->caps.gid_table_len' 3 <= 255
The mlx4_ib_gid_index_to_real_index() does check that "port" is within
bounds, but we don't check the return value for errors. It seems simple
enough to add a check for that.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The story is that Smatch marks skb->data as untrusted so it generates
a warning message here:
drivers/infiniband/hw/cxgb4/cm.c:4100 process_work()
error: buffer overflow 'work_handlers' 241 <= 255
In other places which handle this such as t4_uld_rx_handler() there is
some checking to make sure that the function pointer is not NULL. I
have added bounds checking and a check for NULL here as well.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The pointer reg_workq is local to the source and does not need to be
in global scope, so make it static.
Cleans up sparse warning:
drivers/infiniband/hw/cxgb4/device.c:69:25: warning: symbol 'reg_workq'
was not declared. Should it be static?
Fixes: 1c8f1da5d8 ("iw_cxgb4: Fix possible circular dependency locking warning")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The debugfs file prints the difference between host timestamps as a
seconds/nanoseconds tuple, along with a 64-bit nanoseconds hardware
timestamp. The host time is read using getnstimeofday() which is
deprecated because of the y2038 overflow, and it suffers from time jumps
during settimeofday() and leap seconds.
Converting to ktime_get_ts64() would solve those two, but I'm going
a little further here by changing to ktime_get() and printing 64-bit
nanoseconds on both host and hw timestamps. This simplifies the code
further and makes the output easier to understand.
The format of the debugfs file obviously changes here, but this should
only be read by humans and not scripts, so I assume it's fine.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
With gcc-4.1.2:
drivers/infiniband/core/iwpm_util.c: In function ‘iwpm_send_mapinfo’:
drivers/infiniband/core/iwpm_util.c:647: warning: ‘ret’ may be used uninitialized in this function
Indeed, if nl_client is not found in any of the scanned has buckets, ret
will be used uninitialized.
Preinitialize ret to -EINVAL to fix this.
Fixes: 30dc5e63d6 ("RDMA/core: Add support for iWARP Port Mapper user space service")
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Reviewed-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
If one port fails to initialize an error message should indicate the
reason and driver should continue serving the working port(s) and other
HCA(s).
Fixes: e4b2d06892 ("IB/ipoib: Remove device when one port fails to init").
Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The bnxt_qplib_disable_nq() call is redundant as it occurs
after 'goto fail' and hence it called twice. Remove it.
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
memalloc_noio_save modifies the behavior of MM, we must restore it after
we are done.
Fixes: d83187dda9 ("IB/IPoIB: Convert IPoIB to memalloc_noio_* calls")
Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
There is no need to have a duplication of the generic library, i.e.
hex2bin().
Replace the open coded variant.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
A warning that I thought I had fixed before occasionally comes
back in rare randconfig builds (I found 7 instances in the last
100000 builds, originally it was much more frequent):
drivers/infiniband/hw/mlx5/mr.c: In function 'mlx5_ib_reg_user_mr':
drivers/infiniband/hw/mlx5/mr.c:1229:5: error: 'order' may be used uninitialized in this function [-Werror=maybe-uninitialized]
if (order <= mr_cache_max_order(dev)) {
^
drivers/infiniband/hw/mlx5/mr.c:1247:8: error: 'ncont' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/infiniband/hw/mlx5/mr.c:1247:8: error: 'page_shift' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/infiniband/hw/mlx5/mr.c:1260:2: error: 'npages' may be used uninitialized in this function [-Werror=maybe-uninitialized]
I've looked at all those findings again and noticed that they are all
with CONFIG_INFINIBAND_USER_MEM=n, which means ib_umem_get() returns
an error unconditionally and we never initialize or use those variables.
This triggers a condition in gcc iff mr_umem_get() is partially but not
entirely inlined, which in turn depends on the exact combination of
optimization settings. This is a known problem with gcc, with no
easy solution in the compiler, so this adds another workaround that
should be more reliable than my previous attempt.
Returning an error from mlx5_ib_reg_user_mr() earlier means that we
can completely bypass the logic that caused the warning, the compiler
can now see that the variable is never accessed.
Fixes: 14ab8896f5 ("IB/mlx5: avoid bogus -Wmaybe-uninitialized warning")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
add_sd_cnt in info structure passed to i40iw_create_hmc_obj_type
must be 0 and since it is modified during the call, it must be
reset in the loop. This avoids unnecessarily reprogramming the
SDs multiple times with the same values.
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Use sqsize instead of I40IW_CQP_SW_SQSIZE_2048 to initialize
cqp_requests elements in the for-loop as sqsize is used
to allocate memory for cqp_requests.
Signed-off-by: Chien Tin Tung <chien.tin.tung@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
pr_* is the preferred way to print messages, replace all
printk(KERN_WARN, ...) with pr_warn.
Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Fix ptr_ret.cocci warnings:
drivers/infiniband/core/uverbs_cmd.c:1156:1-3: WARNING: PTR_ERR_OR_ZERO can be used
Use PTR_ERR_OR_ZERO rather than if(IS_ERR(...)) + PTR_ERR
Generated by: scripts/coccinelle/api/ptr_ret.cocci
Signed-off-by: Vasyl Gomonovych <gomonovych@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
There is a stale entry in nes_cm_tcp_context that has apparently
never been used in mainline linux.
I'm trying to kill off all users of timeval as part of the
y2038-safety work, so let's just remove this one.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
There is a stale entry in i40iw_cm_tcp_context that apparently
was copied from the 'nes' driver but never used in i40iw.
I'm trying to kill off all users of timeval as part of the
y2038-safety work, so let's just remove this one.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
There is no need to re-calculate the number of pages since it is already
done in ib_umem_get.
Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Acked-by: Adit Ranadive <aditr@vmware.com>
Tested-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
