Report that driver supports IB_PMA_CLASS_CAP_EXT_WIDTH in respond for
IB_MGMT_CLASS_PERF_MGMT mad with IB_PMA_CLASS_PORT_INFO attr id.
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

When attribute IB_PMA_PORT_COUNTERS_EXT is set, we now return 64 bit
values for the counters.
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Today ocrdma driver defer arming the CQ till poll is called.
This was used to prevent calling poll-cq on an armed CQ.

Recently a set of new CQ API has been introduced into the linux
kernel. The implementation of this API guarantees that a given
CQ is never armed before calling poll on it. Most of the kernel
ULPs have already moved to use this new API or have a code where
poll is called before arming the CQ.

Thus, the above workaround in ocrdma is not needed anymore.
This patch removes the additional logic to deffer arm till poll
is called. This patch adds a simple scheme where ib_req_notify_cq()
will actually arm the cq.
Signed-off-by: NDevesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Fix the RC QPs send queue overhead computation to take into account
two additional segments in the WQE which are needed for registration
operations.

The ATOMIC and UMR segments can't coexist together, so chose maximum out
of them.

The commit 9e65dc37 ("IB/mlx5: Fix RC transport send queue overhead
computation") was intended to update RC transport as commit messages
states, but added the code to UC transport.

Fixes: 9e65dc37 ("IB/mlx5: Fix RC transport send queue overhead computation")
Signed-off-by: NKamal Heib <kamalh@mellanox.com>
Signed-off-by: NLeon Romanovsky <leonro@mellanox.com>
Reviewed-by: NSagi Grimberg <sagig@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

A narrow window for race condition still exist between
multicast join thread and *dev_flush workers.
A kernel crash caused by prolong erratic link state changes
was observed (most likely a faulty cabling):

[167275.656270] BUG: unable to handle kernel NULL pointer dereference at
0000000000000020
[167275.665973] IP: [<ffffffffa05f8f2e>] ipoib_mcast_join+0xae/0x1d0 [ib_ipoib]
[167275.674443] PGD 0
[167275.677373] Oops: 0000 [#1] SMP
...
[167275.977530] Call Trace:
[167275.982225]  [<ffffffffa05f92f0>] ? ipoib_mcast_free+0x200/0x200 [ib_ipoib]
[167275.992024]  [<ffffffffa05fa1b7>] ipoib_mcast_join_task+0x2a7/0x490
[ib_ipoib]
[167276.002149]  [<ffffffff8109d5fb>] process_one_work+0x17b/0x470
[167276.010754]  [<ffffffff8109e3cb>] worker_thread+0x11b/0x400
[167276.019088]  [<ffffffff8109e2b0>] ? rescuer_thread+0x400/0x400
[167276.027737]  [<ffffffff810a5aef>] kthread+0xcf/0xe0
Here was a hit spot:
ipoib_mcast_join() {
..............
      rec.qkey      = priv->broadcast->mcmember.qkey;
                                       ^^^^^^^
.....
 }
Proposed patch should prevent multicast join task to continue
if link state change is detected.
Signed-off-by: NAlex Estrin <alex.estrin@intel.com>

Changes from v4:
- as suggested by Doug Ledford, optimized spinlock usage,
i.e. ipoib_mcast_join() is called with lock held.
Changes from v3:
- sync with priv->lock before flag check.
Chages from v2:
- Move check for OPER_UP flag state to mcast_join() to
ensure no event worker is in progress.
- minor style fixes.
Changes from v1:
- No need to lock again if error detected.
Signed-off-by: NDoug Ledford <dledford@redhat.com>

When checking specific attribute from a bit mask, need to use bitwise
AND and not logical AND, fixed that.

Fixes: 145d9c54 ('IB/core: Display extended counter set if
available')
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: NChristoph Lameter <cl@linux.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

During the ocrdma device remove sequence, the debugfs directory
tree of each ocrdma device needs to be removed. Use
debugfs_remove_recursive instead of debugfs_remove.
Signed-off-by: NSelvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Currently returning the pkey value instead of pkey index.
pkey index is always zero since ocrdma supports only default
pkey.
Signed-off-by: NSelvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

max_sge_rd is used by some of the ULPs to calculate the maximum
number of SGEs that can be used for RDMA READ. Populating this
value in the response of query_device verb. Also, avoid checking
the max_srq_sge while populating max_sge.
Signed-off-by: NSelvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

In the latest kernel, process_mad hook of the driver can be invoked as
soon as device is registered. In this hook, ocrdma driver is issuing a
command to get the stats counters from the HW. This is triggering system
crash since the statistics command resources are not allocated by the driver.
Changing the sequence of initialization to avoid this crash.
Signed-off-by: NSelvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

_show_port_gid_attr performs a va_end on some unused va_list args.
Clean this up by removing the args completely.

Fixes: 470be516 ("IB/core: Add gid attributes to sysfs")
Signed-off-by: NColin Ian King <colin.king@canonical.com>
Reviewed-by: NIra Weiny <ira.weiny@intel.com>
Reviewed-by: NJason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

We are seeing this warning: at net/core/skbuff.c:4174
and before commit a44878d1 ("IB/ipoib: Use one linear skb in RX flow")
skb truesize was not being set when ipoib was using just one skb.
Removing this line avoids the warning when running tcp tests like iperf.

Fixes: a44878d1 ("IB/ipoib: Use one linear skb in RX flow")
Signed-off-by: NCarol L Soto <clsoto@linux.vnet.ibm.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

For GSI QP traffic, the count of the udp header bytes was missing from
the IPv6 header, fix that.

Fixes: 25f40220 ('IB/core: Initialize UD header structure with IP
                     and UDP headers')
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

MLX5_GET64 was used on end_padding_mode, which is a 2-bit field.
This is wrong as the calculated offset is incorrect. Using MLX5_GET
instead of MLX5_GET64 to fix that.

Fixes: 0fb2ed66 ('IB/mlx5: Add create and destroy functionality
                     for Raw Packet QP')
Signed-off-by: NMaor Gottlieb <maorg@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

When a Raw Ethernet QP is created, a NULL pointer PD could be used.
Fixing that by only using the PD after validating it's valid.
smatch also reported this error:
drivers/infiniband/hw/mlx5/qp.c:1629 mlx5_ib_create_qp()
	 error: we previously assumed 'pd' could be null (see line 1616)

Fixes: 0fb2ed66 ('IB/mlx5: Add create and destroy functionality for Raw Packet QP')
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Older libraries that don't have all the new req_v2 fields
should be able to work as well. Today, if the library uses v2, it
will fail to allocate context since the size of reqlen is smaller
than the req_v2 size.

Fix the validation to be with the original req_v2 size and not
the current.

Fixes: f72300c5 ('IB/mlx5: Expose CQE version to user-space')
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

The mlx5_ib driver supports the extended create_cq and create_qp user
verbs. In the current mechanism, a vendor supporting an extended uverb
should set the appropriate bit in the uverbs_ex_cmd_mask field.
Adding the actual support by setting the required bits in order to
support features like completion time-stamping and cross-channel.

Fixes: 972ecb82 ('IB/mlx5: Add create_cq extended command')
Fixes: ddf9529b ('IB/core: Allow setting create flags in QP init
                      attribute')
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

parallel to mutex_{lock,unlock,trylock,is_locked,lock_nested},
inode_foo(inode) being mutex_foo(&inode->i_mutex).

Please, use those for access to ->i_mutex; over the coming cycle
->i_mutex will become rwsem, with ->lookup() done with it held
only shared.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The create_cq() can receive creation flags which were used
differently by two commits which added create_cq extended
command and cross-channel. The merged code caused to not
accept any flags at all.

This patch unifies the check into one function and one return
error code.

Fixes: 972ecb82 ("IB/mlx5: Add create_cq extended command")
Fixes: 051f2630 ("IB/mlx5: Add driver cross-channel support")
Signed-off-by: NLeon Romanovsky <leonro@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Added Raw Packet QP modify functionality which will enable user
space consumers to use it.

Since Raw Packet QP is built of SQ and RQ sub-objects, therefore
Raw Packet QP state changes are implemented by changing the state
of the sub-objects.
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

When modifying a QP, the desired operation was determined in
the mlx5_core using a transition table that takes the current
state, the final state, and returns the desired operation.

Since this logic will be used for Raw Packet QP, move the
operation table to the mlx5_ib.
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

When the user changes the Address Vector(AV) in the modify QP, he
provides an SL. This SL should be translated to Ethernet Priority
by taking the 3 LSB bits, and modify the QP's TIS according to this
Ethernet priority.
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Since Raw Packet QP is composed of RQ and SQ, the IB QP's
state is derived from the sub-objects. Therefore we need
to query each one of the sub-objects, and decide on the
IB QP's state.
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

This patch adds support for Raw Packet QP for the mlx5 device.

Raw Packet QP, unlike other QP types, has no matching mlx5_core_qp
object but rather it is built of RQ/SQ/TIR/TIS/TD mlx5_core object.

Since the SQ and RQ work-queue (WQ) buffers are not contiguous like
other QPs, we allocate separate buffers in the user-space and pass
the address of each one of them separately to the kernel.
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Extract specific IB QP fields to mlx5_ib_qp_trans structure.
The mlx5_core QP object resides in mlx5_ib_qp_base, which all QP types
inherit from. When we need to find mlx5_ib_qp using mlx5_core QP
(event handling and co), we use a pointer that resides in
mlx5_ib_qp_base.

In addition, we delete all redundant fields that weren't used anywhere
in the code:
-doorbell_qpn
-sq_max_wqes_per_wr
-sq_spare_wqes
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Transport Domain groups several TIS and TIR object. By grouping
these object, it defines wheather local loopback packets that
are sent from the TIS objects in the group are received by the
TIR objects in the same group.

Allocate a Transport Domain(TD) for each user context to be used
in the future by Raw Packet QP for Self-Loopback Control.
Signed-off-by: NMajd Dibbiny <majd@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Per user context, work with CQE version that both the user-space
and the kernel support. Report this CQE version via the response of
the alloc_ucontext command.
Signed-off-by: NHaggai Abramovsky <hagaya@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Enforce working with CQE version 1 when the user supports CQE
version 1 and asked to work this way.

If the user still works with CQE version 0, then use the default
CQE version to tell the Firmware that the user still works in the
older mode.

After this patch, the kernel still reports CQE version 0.
Signed-off-by: NHaggai Abramovsky <hagaya@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

The wrong buffer size was passed to ib_is_udata_cleared.
Signed-off-by: NHaggai Abramovsky <hagaya@mellanox.com>
Reviewed-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

The rdma netlink local service registers a handler to handle RESOLVE
response and another handler to handle SET_TIMEOUT request. The first
thing these handlers do is to call netlink_capable() to check the
access right of the received skb to make sure that the sender has root
access. Under normal conditions, such responses and requests will be
directly forwarded to the handlers without going through the netlink_dump
pathway (see ibnl_rcv_msg() in drivers/infiniband/core/netlink.c).
However, a user application could send a RESOLVE request (not response)
to the local service, which will fall into the netlink_dump pathway,
where a new skb will be created without initializing the control block.
This new skb will be eventually forwarded to the local service RESOLVE
response handler. Unfortunately, netlink_capable() will cause general
protection fault if the skb's control block is not initialized. This
patch will address the problem by checking the skb first.
Signed-off-by: NKaike Wan <kaike.wan@intel.com>
Reported-by: NDmitry Vyukov <dvyukov@google.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

No usage after the conversion to the new CQ API.
Signed-off-by: NSagi Grimberg <sagig@mellanox.com>
Reviewed-by: NBart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Based on profiling, UD performance drops in case of processes
in a single client due to excess context switches when
the progress workqueue is scheduled.

This is solved by modifying the heuristic to select the
direct progress instead of the scheduling progress via
the workqueue when UD-like situations are detected in
the heuristic.
Reviewed-by: NVinit Agnihotri <vinit.abhay.agnihotri@intel.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Advertise RoCE v2 support in port_immutable attributes according to
the hardware's capabilities. This enables the verbs stack to use
RoCE v2 mode.
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

The mlx4 driver uses a special QP to implement the GSI QP. This kind
of QP allows to build the InfiniBand headers in software.
When mlx4 hardware builds the packet, it calculates the ICRC and puts
it at the end of the payload. However, this ICRC calculation depends
on the QP configuration, which is determined when the QP is modified
(roce_mode during INIT->RTR).
When receiving a packet, the ICRC verification doesn't depend on this
configuration.
Therefore, using two GSI QPs for send (one for each RoCE version) and
one GSI QP for receive are required.
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

RoCEv2 packets are sent over IP/UDP protocols.
The mlx4 driver uses a type of RAW QP to send packets for QP1 and
therefore needs to build the network headers below BTH in software.

This patch adds option to build QP1 packets with IP and UDP headers if
RoCEv2 is requested.
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

If the hardware supports RoCE v2, we configure the hardware UDP
port according to the RoCE v2 Annex when mlx4_ib device is added.
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

In order to support modify_qp for RoCE v2, we need to set
the gid_type in the QP context.
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

To tell hardware about a gid with type RoCEv2, software needs a new
modifier to the SET_PORT command: MLX4_SET_PORT_ROCE_ADDR. This can
replace the old method, MLX4_SET_PORT_GID_TABLE, for  RoCEv1 gids.
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

IB core driver adds a property of type to struct ib_gid_attr.
The mlx4 driver should take that in consideration when modifying or
querying the hardware gid table.
Signed-off-by: NMoni Shoua <monis@mellanox.com>
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>

Previously, IPV6_DEFAULT_HOPLIMIT was used as the hop limit value for
RoCE. Fixing that by taking ip4_dst_hoplimit and ip6_dst_hoplimit as
hop limit values.
Signed-off-by: NMatan Barak <matanb@mellanox.com>
Signed-off-by: NDoug Ledford <dledford@redhat.com>