IB/hfi1: Do not allocate PIO send contexts for VNIC

OPA VNIC does not use PIO contexts and instead only uses SDMA
engines. Do not allocate PIO contexts for VNIC ports.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Remove unnecessary if check

A for loop condition of data_iovs in user_sdma_free_request
is unnecessarily repeated before the loop as an if check.

Remove the if enveloping the loop.

Reviewed-by: Jakub Byczkowski <jakub.byczkowski@intel.com>
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix a wrapping test to insure the correct timeout

The "2 * UINT_MAX"  statement:
if ((u64)(ts - cce->timestamp) > 2 * UINT_MAX) {

is equivalent to:
if ((u64)(ts - cce->timestamp) > UINT_MAX - 1) {

This results in a premature timeout of the cong log entry.

Fix by using unsigned 64 bit integers, removing casts, and using
an algebraic equivalent test to avoid the "2 * UINT_MAX" issue.

Also make use of kernel API to get nanoseconds instead of
open coding.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Remove wrapper function in mmu_rb

Wrapper functions were used to call the same function
mmu_notifier_mem_invalidate for 2 callbacks in
mmu_notifier.
The commit 7def96f0a973 ("IB/hfi1: update to new mmu_notifier semantic")
removed the invalidate_page callback.
Therefore, the wrapper function is no longer needed.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Reduce 8051 command timeout

Timeout of 20 seconds is too long for active wait performed
for 8051 command completion. It was required for scenarios
when transition to polling was requested before offline.quiet
state was reached. Currently wait for offline.quiet is
properly implemented and timeout can be reduced to 1 second.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Duane McCrory <duane.mccrory@intel.com>
Signed-off-by: Jakub Byczkowski <jakub.byczkowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Allow MgmtAllowed on B2B setups

HFI's are hard-wired to send Device Info frames with
MgmtAllowed bit set to 0. This means in B2B setups,
MgmtAllowed would never be allowed, which prevents
remote opa management tools from working properly.

Assume MgmtAllowed if a neighbor is also an HFI.

Fixes: 98b9ee2002a8 ("IB/hfi1: Cache neighbor secure data after link up")
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/srpt: Ensure that modifying the use_srq configfs attribute works

The use_srq configfs attribute is created after it is read. Hence
modify srpt_tpg_attrib_use_srq_store() such that this function
switches dynamically between non-SRQ and SRQ mode.

Reported-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/srpt: Wait until channel release has finished during module unload

Introduce the helper function srpt_set_enabled(). Protect
sport->enabled changes with sdev->mutex. Makes configfs writes
into 'enabled' wait until all channel resources have been freed.
Wait until channel release has finished during kernel module
unload.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/srpt: Introduce srpt_disconnect_ch_sync()

Except for changing a BUG_ON() call into a WARN_ON_ONCE() call, this
patch does not change any functionality.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/srpt: Introduce helper functions for SRQ allocation and freeing

The only functional change in this patch is in the srpt_add_one()
error path: if allocating the ring buffer for the SRQ fails, fall
back to non-SRQ mode instead of disabling SRP target functionality.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/srpt: Post receive work requests after qp transition to INIT state

IB and iWARP specs both spell out that posting a receive work request
to a queue pair in the RESET state is an invalid operation and required
to fail.  Postpone posting receive work requests until after the
transition to the INIT state.

Fixes: commit dea262094cdf ("IB/srpt: Change default behavior from using SRQ to using RC")
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

iw_cxgb4: remove BUG_ON() usage.

iw_cxgb4 has many BUG_ON()s that were left over from various enhancemnets
made over the years.  Almost all of them should just be removed.  Some,
however indicate a ULP usage error and can be handled w/o bringing down
the system.

If the condition cannot happen with correctly implemented cxgb4 sw/fw,
then remove the BUG_ON.

If the condition indicates a misbehaving ULP (like CQ overflows), add
proper recovery logic.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

bnxt_re: changing the ip address shouldn't affect new connections

While adding a new gid, the driver currently does not return the context
back to the stack. A subsequent del_gid() (e.g, when ip address is changed)
doesn't find the right context in the driver and it ends up dropping that
request. This results in the HW caching a stale gid entry and traffic fails
because of that. Fix by returning the proper context in bnxt_re_add_gid().

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

bnxt_re: fix a crash in qp error event processing

In bnxt_qplib_process_qp_event(), for qp error events we look up the
qp-handle and pass it for further processing. But we don't check if the
handle is NULL. This could lead to a crash in the called functions when
that qp-handle is dereferenced, if the qp is destroyed in the meantime.
Fix this by checking for a valid qp-handle in that function.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Fix ABI alignment to 64 bit

Struct mlx5_ib_striding_rq_caps was not aligned to 64 bit as
it should have been. Add a 32 bit reserved field.

Fixes: b4f34597a5ce ('IB/mlx5: Expose multi-packet RQ capabilities')
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Avoid unnecessary return value check

Since there is nothing done with non zero return value, such check is
avoided.

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: Doug Ledford <dledford@redhat.com>

IB/mlx4: Increase maximal message size under UD QP

Maximal message should be used as a limit to the max message payload allowed,
without the headers. The ConnectX-3 check is done against this value includes
the headers. When the payload is 4K this will cause the NIC to drop packets.

Increase maximal message to 8K as workaround, this shouldn't change current
behaviour because we continue to set the MTU to 4k.

To reproduce;
set MTU to 4296 on the corresponding interface, for example:
ifconfig eth0 mtu 4296 (both server and client)

On server:
ib_send_bw -c UD -d mlx4_0 -s 4096 -n 1000000 -i1 -m 4096

On client:
ib_send_bw -d mlx4_0 -c UD <server_ip> -s 4096 -n 1000000 -i 1 -m 4096

Fixes: 6e0d733d9215 ("IB/mlx4: Allow 4K messages for UD QPs")
Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Add contig support for control objects

Taking advantage of the optimization which was introduced in previous
commit ("IB/mlx4: Use optimal numbers of MTT entries") to optimize the
MTT usage for QP and CQ.

Signed-off-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Use optimal numbers of MTT entries

Optimize the device performance by assigning multiple physical pages,
which are contiguous, to a single MTT. As a result, the number of MTTs
is reduced and in turn save cache misses of MTTs.

Signed-off-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Fix RoCE Address Path fields

When working over a RoCE network, the UDP source port should be set only
for statically connected QPs (RC, UC and XRC).

Fixes: 2811ba51b049 ("IB/mlx5: Add RoCE fields to Address Vector")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Assign send CQ and recv CQ of UMR QP

The UMR's QP is created by calling mlx5_ib_create_qp directly, and
therefore the send CQ and the recv CQ on the ibqp weren't assigned.

Assign them right after calling the mlx5_ib_create_qp to assure
that any access to those pointers will work as expected and won't
crash the system as might happen as part of reset flow.

Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/cxgb4: Protect from possible dereference

Smatch tool reports the following error:
  drivers/infiniband/hw/cxgb4/qp.c:1886
c4iw_create_qp() error: we previously assumed 'ucontext'
could be null (see line 1804)

Cc: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/bnxt_re: Remove unused vlan_tag variable

The Broadcom driver produces the following compilation warning

drivers/infiniband/hw/bnxt_re/ib_verbs.c:
In function ‘bnxt_re_create_ah’:
drivers/infiniband/hw/bnxt_re/ib_verbs.c:668:6:
warning: variable ‘vlan_tag’ set but not used [-Wunused-but-set-variable]
u16 vlan_tag;

Let's remove it till vlan_tag will be implemented properly.

Cc: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Add PCI write end padding support

Add the PCI write end padding flag to device_cap_flags enum and set it
during mlx5_ib_query_device so it will be reported to user-space.

During WQ/QP creation, set that capability for WQ/QP if user requested
it and HW supports it.

PCI write end padding modification is not supported for now. There's no
such flag for a QP but for a WQ, create and modify use the same flag.
Return an error if PCI write end padding flag is set during modify_wq.

Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Add PCI write end padding flags for WQ and QP

There are root complexes that are able to optimize their
performance when incoming data is multiple full cache lines.

PCI write end padding is the device's ability to pad the ending of
incoming packets (scatter) to full cache line such that the last
upstream write generated by an incoming packet will be a full cache
line.

Add a relevant entry to ib_device_cap_flags to report such capability
of an RDMA device.

Add the QP and WQ create flags:
 * A QP/WQ created with a scatter end padding flag will cause
   HW to pad the last upstream write generated by a packet to cache line.

User should consider several factors before activating this feature:
- In case of high CPU memory load (which may cause PCI back pressure in
  turn), if a large percent of the writes are partial cache line, this
  feature should be checked as an optional solution.
- This feature might reduce performance if most packets are between one
  and two cache lines and PCIe throughput has reached its maximum
  capacity. E.g. 65B packet from the network port will lead to 128B
  write on PCIe, which may cause traffic on PCIe to reach high
  throughput.

Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: don't crash, if allocation of crc algorithm failed

Following crash happens, if crc algorithm couldn't be allocated:

[ 1087.989072] rdma_rxe: loaded
[ 1097.855397] PCLMULQDQ-NI instructions are not detected.
[ 1097.901220] rdma_rxe: failed to allocate crc algorithmi err:-2
[ 1097.901248] BUG: unable to handle kernel
[ 1097.901249] NULL pointer dereference
[ 1097.901250]  at 0000000000000046
[...]

Reason is that rxe->tfm is assigned the error return, which will then
be used for crypto_free_shash() in rxe_cleanup. Fix by using a
temporary variable and assigning it rxe->tfm after allocation succeeded.

Fixes: cee2688e3cd6 ("IB/rxe: Offload CRC calculation when possible")
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Avoid crash on pkey enforcement failed in received MADs

Below kernel crash is observed when Pkey security enforcement fails on
received MADs. This issue is reported in [1].

ib_free_recv_mad() accesses the rmpp_list, whose initialization is
needed before accessing it.
When security enformcent fails on received MADs, MAD processing avoided
due to security checks failed.

OpenSM[3770]: SM port is down
kernel: BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
kernel: IP: ib_free_recv_mad+0x44/0xa0 [ib_core]
kernel: PGD 0
kernel: P4D 0
kernel:
kernel: Oops: 0002 [#1] SMP
kernel: CPU: 0 PID: 2833 Comm: kworker/0:1H Tainted: P          IO    4.13.4-1-pve #1
kernel: Hardware name: Dell       XS23-TY3        /9CMP63, BIOS 1.71 09/17/2013
kernel: Workqueue: ib-comp-wq ib_cq_poll_work [ib_core]
kernel: task: ffffa069c6541600 task.stack: ffffb9a729054000
kernel: RIP: 0010:ib_free_recv_mad+0x44/0xa0 [ib_core]
kernel: RSP: 0018:ffffb9a729057d38 EFLAGS: 00010286
kernel: RAX: ffffa069cb138a48 RBX: ffffa069cb138a10 RCX: 0000000000000000
kernel: RDX: ffffb9a729057d38 RSI: 0000000000000000 RDI: ffffa069cb138a20
kernel: RBP: ffffb9a729057d60 R08: ffffa072d2d49800 R09: ffffa069cb138ae0
kernel: R10: ffffa069cb138ae0 R11: ffffa072b3994e00 R12: ffffb9a729057d38
kernel: R13: ffffa069d1c90000 R14: 0000000000000000 R15: ffffa069d1c90880
kernel: FS:  0000000000000000(0000) GS:ffffa069dba00000(0000) knlGS:0000000000000000
kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
kernel: CR2: 0000000000000008 CR3: 00000011f51f2000 CR4: 00000000000006f0
kernel: Call Trace:
kernel:  ib_mad_recv_done+0x5cc/0xb50 [ib_core]
kernel:  __ib_process_cq+0x5c/0xb0 [ib_core]
kernel:  ib_cq_poll_work+0x20/0x60 [ib_core]
kernel:  process_one_work+0x1e9/0x410
kernel:  worker_thread+0x4b/0x410
kernel:  kthread+0x109/0x140
kernel:  ? process_one_work+0x410/0x410
kernel:  ? kthread_create_on_node+0x70/0x70
kernel:  ? SyS_exit_group+0x14/0x20
kernel:  ret_from_fork+0x25/0x30
kernel: RIP: ib_free_recv_mad+0x44/0xa0 [ib_core] RSP: ffffb9a729057d38
kernel: CR2: 0000000000000008

[1] : https://www.spinics.net/lists/linux-rdma/msg56190.html

Fixes: 47a2b338fe63 ("IB/core: Enforce security on management datagrams")
Cc: stable@vger.kernel.org # 4.13+
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reported-by: Chris Blake <chrisrblake93@gmail.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Reviewed-by: Hal Rosenstock <hal@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/cxgb4: Annotate r2 and stag as __be32

Chelsio cxgb4 HW is big-endian, hence there is need to properly
annotate r2 and stag fields as __be32 and not __u32 to fix the
following sparse warnings.

  drivers/infiniband/hw/cxgb4/qp.c:614:16:
    warning: incorrect type in assignment (different base types)
      expected unsigned int [unsigned] [usertype] r2
      got restricted __be32 [usertype] <noident>
  drivers/infiniband/hw/cxgb4/qp.c:615:18:
    warning: incorrect type in assignment (different base types)
      expected unsigned int [unsigned] [usertype] stag
      got restricted __be32 [usertype] <noident>

Cc: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Fix RSS's QPC attributes assignments

In the modify QP handler the base_qpn_udp field in the RSS QPC is
overwrite later by irrelevant value assignment. Hence, ingress packets
which gets to the RSS QP will be steered then to a garbage QPN.

The patch fixes this by skipping the above assignment when a RSS QP is
modified, also, the RSS context's attributes assignments are relocated
just before the context is posted to avoid future issues like this.

Additionally, this patch takes the opportunity to change the code to be
disciplined to the device's manual and assigns the RSS QP context just at
RESET to INIT transition.

Fixes:3078f5f1bd8b ("IB/mlx4: Add support for RSS QP")
Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Add report for RSS capabilities by vendor channel

The mlx4's RSS patches submission missed a report of RSS capabilities
which should be reported by the vendor channel in query_device.

Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/umem: Avoid partial declaration of non-static function

The RDMA/umem uses generic RB-trees macros to generate various ib_umem
access functions. The generation is performed with INTERVAL_TREE_DEFINE
macro, which allows one of two modes: declare all functions as static or
declare none of the function to be static.

The second mode of operation produces the following sparse errors:
 drivers/infiniband/core/umem_rbtree.c:69:1:
warning: symbol 'rbt_ib_umem_iter_first' was not declared.
Should it be static?
 drivers/infiniband/core/umem_rbtree.c:69:1:
warning: symbol 'rbt_ib_umem_iter_next' was not declared.
Should it be static?

Code relocation together with declaration of such functions to be
"static" solves the issue.

Because there is no need to have separate file for two functions,
let's consolidate umem_rtree.c and umem_odp.c into one file.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Modify the usage of cmd_sn in hip08

The cmd_sn field of CQ doorbell inits for 0. It should be
increment on each first db rung after a completion Event.
if the cmd_sn of notify doorbell Adjacent two times is the
same, the hardware will distinguish it for the same notify
request and update its type according to the priority level
of next event and solicited event.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Unify the calculation for hem index in hip08

The calculation of hem index are different between hns_roce_table_get
and hns_roce_table_find. When the table chunk size of TRRL is not
divisible by object size, it will faile to find the trrl table.

This patch is to update the calculation of the hem index in the
hns_roce_table_find to the same as which in the hns_roce_table_get.

Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Set the owner field of SQWQE in hip08 RoCE

the owner need to be set when posting sqwqe in hip08 RoCE.
The owner be used according to the below algorithm:
The value of owner should be 1 in the first lap, it
should be 0 in the second lap and in turn.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Add sq_invld_flg field in QP context

In hip08 RoCE, it need to add the sq_invld_flg field
in QP context for RoCE hardware.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Update the usage of ack timeout in hip08

The ack timeout's value in qp context shall be a 5-bit value
and be assgined by users. When at of qpc is set zero, the
timer is disabled.

When attr_mask set for IB_QP_TIMEOUT, The ack timeout field
is effective.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Set sq_cur_sge_blk_addr field in QPC in hip08

If the extend sges exist, the sq_cur_sge_blk_addr field in QPC
(qp context) should be configured.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Enable the cqe field of sqwqe of RC

When sig_type of qpc is non-selectable, all sq's wqes will produce
cqe and not depend on the cqe attribute of wqe. When sig_type of
qpc is selectable, The cqe attribute of wqe will decide whether to
produce the cqe.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Set se attribute of sqwqe in hip08

When send flags is IB_SEND_SOLICITED, the se(solicated event)
field of sqwqe will be set.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Configure fence attribute in hip08 RoCE

When post wr for mixed rdma operation, we need to use fence
mechanism to keep the correct execute order.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Configure TRRL field in hip08 RoCE device

The TRRL(Target RDMA Read/aTOMIC List) record the information
of receiving RDMA READ or ATOMIC operation in hip08. It will
be used the hardware. The driver need to assign a continuous
physical address for trrl_ba field of qp context.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Update calculation of irrl_ba field for hip08

The irrl(initiator RDMA Read/Atomic list) base address of qp
context is assigned for addr[63:6]. This patch mainly fixed
it.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Configure sgid type for hip08 RoCE

The hardware vendors need to generate RoCEv1 or RoCEv2
packet according to the sgid type configured.

Besides, update the gid table size for hip08 RoCE
device.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Generate gid type of RoCEv2

HNS_ROCE_CAP_FALG_ROCE_V1_V2 is added for selecting capability of
RoCE in hns driver. When HNS_ROCE_CAP_FALG_ROCE_V1_V2 is set,
driver will inform ib core that the related hns device can support
RoCEv2, and ib core can generate the gid of the related type.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Add rereg mr support for hip08

This patch adds rereg mr support for hip08.

Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge branch 'k.o/for-rc' into k.o/for-next

Pick up the missing netlink oops fix

Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Take advantage of kvzalloc_node in sdma initialization

The code that allocates the tx ring in the sdma code fails to take
advantage of kvzalloc variations.

Fix by converting to use kvzalloc_node.

Reported-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Don't modify num_user_contexts module parameter

The driver parameter num_user_contexts controls global behavior and
should not be modified by the driver.
This patch eliminates modification of num_user_contexts by using a
local variable to keep track of the value.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Insure int mask for in-kernel receive contexts is clear

The only use for the urg interrupt is for priority PSM packets.

There is no reason for this interrupt to be enabled for kernel
contexts.

Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add tx_opcode_stats like the opcode_stats

This patch adds tx_opcode_stats to parallel the
(rx)opcode_stats in the debugfs.

Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Validate PKEY for incoming GSI MAD packets

These are the use-cases where the pkey needs to be tested to see
if a packet needs to be dropped.

a) Check if pkey is not FULL_MGMT_P_KEY or LIM_MGMT_P_KEY,
   drop the packet as it's not part of the management partition.
   Self-originated packets are an exception.

b) If pkey index points to FULL_MGMT_P_KEY and LIM_MGMT_P_KEY is
   in the table, the packet is coming from a management node,
   and the receiving node is also a management node, so it is safe
   for the packet to go through.

c) If pkey index points to FULL_MGMT_P_KEY and LIM_MGMT_P_KEY is
   NOT in the table, drop the packet as LIM_MGMT_P_KEY should
   always be in the pkey table. It could be a misconfiguration.

d) If pkey index points to LIM_MGMT_P_KEY and FULL_MGMT_P_KEY is
   NOT in the table, it is safe for the packet to go through
   since a non-management node is talking to another non-managment
   node.

e) If pkey index points to LIM_MGMT_P_KEY and FULL_MGMT_P_KEY is in
   the table, drop the packet because a non-management node is
   talking to a management node, and it could be an attack.

For the implementation, these rules can be simplied to only checking
for (a) and (e). There's no need to check for rule (b) as
the packet doesn't need to be dropped. Rule (c) is not possible in
the driver as LIM_MGMT_P_KEY is always in the pkey table.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Ib/hfi1: Return actual operational VLs in port info query

__subn_get_opa_portinfo stores value returned by hfi1_get_ib_cfg() as
operational vls. hfi1_get_ib_cfg() returns vls_operational field in
hfi1_pportdata. The problem with this is that the value is always equal
to vls_supported field in hfi1_pportdata.

The logic to calculate operational_vls is to set value passed by FM
(in  __subn_set_opa_portinfo routine). If no value is passed then
default value is stored in operational_vls.

Field actual_vls_operational is calculated on the basis of buffer
control table. Hence, modifying hfi1_get_ib_cfg() to return
actual_operational_vls when used with HFI1_IB_CFG_OP_VLS parameter

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Patel Jay P <jay.p.patel@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Race condition between user notification and driver state

The handler for link init state (HLS_UP_INIT) notifies userspace
(update_statusp()) before enabling the device
(RCV_CTRL_RCV_PORT_ENABLE_SMASK) or setting the device state
(ppd->host_link_state).  This causes a race condition where the
userspace thinks the interface is in the INIT state before the driver
has set that state.

Rework the code path to eliminate the race.

Delay setting the init state until after a HW settling period.

Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

bnxt_re: Implement the shutdown hook of the L2-RoCE driver interface

When host is shutting down, it invokes the shutdown hook of the
L2 driver where it would attempt to free the MSI-X vectors, but would fail
because some vectors are held by the RoCE driver.
Implement the new hook in the L2 -> RoCE interface which will be invoked so that
the RoCE driver can unregister the device and free up the MSI-X vectors it had
claimed so that L2 can proceed with it's shutdown without failure.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/cxgb4: Declare stag as __be32

The scqe.stag is actually __b32, fix it.

  drivers/infiniband/hw/cxgb4/cq.c:754:52: warning: cast to restricted __be32

Cc: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Convert timers to use timer_setup()

In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly.

Cc: Moni Shoua <monis@mellanox.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/netlink: OOPs in rdma_nl_rcv_msg() from misinterpreted flag

rdma_nl_rcv_msg() checks to see if it should use the .dump() callback
or the .doit() callback.  The check is done with this check:

if (flags & NLM_F_DUMP) ...

The NLM_F_DUMP flag is two bits (NLM_F_ROOT | NLM_F_MATCH).

When an RDMA_NL_LS message (response) is received, the bit used for
indicating an error is the same bit as NLM_F_ROOT.

NLM_F_ROOT == (0x100) == RDMA_NL_LS_F_ERR.

ibacm sends a response with the RDMA_NL_LS_F_ERR bit set if an error
occurs in the service.  The current code then misinterprets the
NLM_F_DUMP bit and trys to call the .dump() callback.

If the .dump() callback for the specified request is not available
(which is true for the RDMA_NL_LS messages) the following Oops occurs:

[ 4555.960256] BUG: unable to handle kernel NULL pointer dereference at
   (null)
[ 4555.969046] IP:           (null)
[ 4555.972664] PGD 10543f1067 P4D 10543f1067 PUD 1033f93067 PMD 0
[ 4555.979287] Oops: 0010 [#1] SMP
[ 4555.982809] Modules linked in: rpcrdma ib_isert iscsi_target_mod
target_core_mod ib_iser libiscsi scsi_transport_iscsi ib_ipoib rdma_ucm ib_ucm
ib_uverbs ib_umad rdma_cm ib_cm iw_cm dm_mirror dm_region_hash dm_log dm_mod
dax sb_edac x86_pkg_temp_thermal intel_powerclamp coretemp kvm irqbypass
crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel crypto_simd
glue_helper cryptd hfi1 rdmavt iTCO_wdt iTCO_vendor_support ib_core mei_me
lpc_ich pcspkr mei ioatdma sg shpchp i2c_i801 mfd_core wmi ipmi_si ipmi_devintf
ipmi_msghandler acpi_power_meter acpi_pad nfsd auth_rpcgss nfs_acl lockd grace
sunrpc ip_tables ext4 mbcache jbd2 sd_mod mgag200 drm_kms_helper syscopyarea
sysfillrect sysimgblt fb_sys_fops ttm igb ahci crc32c_intel ptp libahci
pps_core drm dca libata i2c_algo_bit i2c_core
[ 4556.061190] CPU: 54 PID: 9841 Comm: ibacm Tainted: G          I
4.14.0-rc2+ #6
[ 4556.069667] Hardware name: Intel Corporation S2600WT2/S2600WT2, BIOS
SE5C610.86B.01.01.0008.021120151325 02/11/2015
[ 4556.081339] task: ffff880855f42d00 task.stack: ffffc900246b4000
[ 4556.087967] RIP: 0010:          (null)
[ 4556.092166] RSP: 0018:ffffc900246b7bc8 EFLAGS: 00010246
[ 4556.098018] RAX: ffffffff81dbe9e0 RBX: ffff881058bb1000 RCX:
0000000000000000
[ 4556.105997] RDX: 0000000000001100 RSI: ffff881058bb1320 RDI:
ffff881056362000
[ 4556.113984] RBP: ffffc900246b7bf8 R08: 0000000000000ec0 R09:
0000000000001100
[ 4556.121971] R10: ffff8810573a5000 R11: 0000000000000000 R12:
ffff881056362000
[ 4556.129957] R13: 0000000000000ec0 R14: ffff881058bb1320 R15:
0000000000000ec0
[ 4556.137945] FS:  00007fe0ba5a38c0(0000) GS:ffff88105f080000(0000)
knlGS:0000000000000000
[ 4556.147000] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 4556.153433] CR2: 0000000000000000 CR3: 0000001056f5d003 CR4:
00000000001606e0
[ 4556.161419] Call Trace:
[ 4556.164167]  ? netlink_dump+0x12c/0x290
[ 4556.168468]  __netlink_dump_start+0x186/0x1f0
[ 4556.173357]  rdma_nl_rcv_msg+0x193/0x1b0 [ib_core]
[ 4556.178724]  rdma_nl_rcv+0xdc/0x130 [ib_core]
[ 4556.183604]  netlink_unicast+0x181/0x240
[ 4556.187998]  netlink_sendmsg+0x2c2/0x3b0
[ 4556.192392]  sock_sendmsg+0x38/0x50
[ 4556.196299]  SYSC_sendto+0x102/0x190
[ 4556.200308]  ? __audit_syscall_entry+0xaf/0x100
[ 4556.205387]  ? syscall_trace_enter+0x1d0/0x2b0
[ 4556.210366]  ? __audit_syscall_exit+0x209/0x290
[ 4556.215442]  SyS_sendto+0xe/0x10
[ 4556.219060]  do_syscall_64+0x67/0x1b0
[ 4556.223165]  entry_SYSCALL64_slow_path+0x25/0x25
[ 4556.228328] RIP: 0033:0x7fe0b9db2a63
[ 4556.232333] RSP: 002b:00007ffc55edc260 EFLAGS: 00000293 ORIG_RAX:
000000000000002c
[ 4556.240808] RAX: ffffffffffffffda RBX: 0000000000000010 RCX:
00007fe0b9db2a63
[ 4556.248796] RDX: 0000000000000010 RSI: 00007ffc55edc280 RDI:
000000000000000d
[ 4556.256782] RBP: 00007ffc55edc670 R08: 00007ffc55edc270 R09:
000000000000000c
[ 4556.265321] R10: 0000000000000000 R11: 0000000000000293 R12:
00007ffc55edc280
[ 4556.273846] R13: 000000000260b400 R14: 000000000000000d R15:
0000000000000001
[ 4556.282368] Code:  Bad RIP value.
[ 4556.286629] RIP:           (null) RSP: ffffc900246b7bc8
[ 4556.293013] CR2: 0000000000000000
[ 4556.297292] ---[ end trace 8d67abcfd10ec209 ]---
[ 4556.305465] Kernel panic - not syncing: Fatal exception
[ 4556.313786] Kernel Offset: disabled
[ 4556.321563] ---[ end Kernel panic - not syncing: Fatal exception
[ 4556.328960] ------------[ cut here ]------------

Special case RDMA_NL_LS response messages to call the appropriate
callback.

Additionally, make sure that the .dump() callback is not NULL
before calling it.

Fixes: 647c75ac59a48a54 ("RDMA/netlink: Convert LS to doit callback")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Reviewed-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/cm: Fix memory corruption in handling CM request

In recent code, two path record entries are alwasy cleared while
allocated could be either one or two path record entries.
This leads to zero out of unallocated memory.

This fix initializes alternative path record only when alternative path
is set.

While we are at it, path record allocation doesn't check for OPA
alternative path, but rest of the code checks for OPA alternative path.
Path record allocation code doesn't check for OPA alternative LID.
This can further lead to memory corruption when only one path record is
allocated, but there is actually alternative OPA path record present in CM
request.

Cc: <stable@vger.kernel.org> # v4.12+
Fixes: 9fdca4da4d8c ("IB/SA: Split struct sa_path_rec based on IB and ROCE specific fields")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Add support for RSS on the inner packet

Some user space application would like to do RSS on the inner
packet fields instead on the outer.
When MLX5_RX_HASH_INNER is set with one or more of the other
hash fields, then the RSS will be done using the inner packet.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Add tunneling offloads support

The device can support receive Stateless Offloads for the inner
packet's fields only when the packet is processed by TIR which is
enabled to support tunneling. Otherwise, the device treats the
packet as an ordinary non-tunneling packet and receive offloads
can be done only for the outer packet's field.
In order to enable receive Stateless Offloading support for incoming
tunneling traffic the TIR should be created with tunneled_offload_en.
Tunneling offloads is supported only be raw ethernet QP.

This patch includes:
* New QP creation flag for tunneling offloads.
* Reports device capabilities.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Update tunnel offloads bits

This patch updates the mlx5_ifc with the following:
- Fix tunnel_stateless_gre typo.
- max_geneve_opt_len - Maximum geneve options length.
- tunnel_stateless_geneve_rx - If set, receive Stateless
  Offloads for Geneve tunneled (inner) packets are supported.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Support padded 128B CQE feature

In some benchmarks and some CPU architectures, writing the CQE on a full
cache line size improves performance by saving memory access operations
(read-modify-write) relative to partial cache line change. This patch
lets the user to configure the device to pad the CQE up to 128B in case
its content is less than 128B. Currently the driver supports only padding
for a CQE size of 128B.

Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Support 128B CQE compression feature

In commit 1cbe6fc86ccf ("IB/mlx5: Add support for CQE compressing") the
concept of CQE compression was introduced and added a support for 64B
CQE size. This change update the code to support 128B CQE size as well.

Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Add 128B CQE compression and padding HW bits

Adding new bits in mlx5_ifc_cmd_hca_cap to get the hardware capabilities
for:
 - compression_128: Support 128B CQE compression
 - cqe_128_always: Support 128B CQE padding

Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Allow creation of a multi-packet RQ

Allow creation of a multi-packet receive queue.

In order to create a multi-packet RQ, the following fields in
the mlx5_ib_rwq should be set:
- log_num_strides: Log of number of strides per WQE
- single_stride_log_num_of_bytes: Log of a single stride size
- two_byte_shift_en: When enabled, hardware pads 2 bytes of zeros
  before writing the message to memory (e.g. for the IP alignment).

Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Expose multi-packet RQ capabilities

This patch reports the device's striding RQ capabilities to
the user-space:
- min/max_single_stride_log_num_of_bytes: Log of min/max number of
  bytes in a single stride.
- min/max_single_wqe_log_num_of_strides: Log of min/max number of
  strides in a single WQE.
- supported_qpts: A bit mask to know which QP types support multi-
  packet RQ, for now only Raw Packet QPs.

Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Add modify CQ support for hip08

It is needed to call modify cq API for modifying cq
context fields for controlling event generation
moderations. This patch mainly adds it.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Update the PD&CQE&MTT specification in hip08

This patch updates the PD specification to 16M for hip08. And it
updates the numbers of mtt and cqe segments for the buddy.

As the CQE supports hop num 1 addressing, the CQE specification is
64k. This patch updates to set the CQE specification to 64k.

Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Update the IRRL table chunk size in hip08

As the increase of the IRRL specification in hip08, the IRRL table
chunk size needs to be updated.
This patch updates the IRRL table chunk size to 256k for hip08.

Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Support WQE/CQE/PBL page size configurable feature in hip08

This patch updates to support WQE, CQE and PBL page size configurable
feature, which includes base address page size and buffer page size.

Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/ipoib: Change number of TX wqe to 64

NAPI budget is 64 packets, while maximum polling size for
the send CQ is 16. Let's bring them in sync, so the NAPI
budget will be reused completely.

Cc: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/ipoib: Use NAPI in UD/TX flows

Instead of explicit call to poll_cq of the tx ring, use the NAPI mechanism
to handle the completions of each packet that has been sent to the HW.

The next major changes were taken:
 * The driver init completion function in the creation of the send CQ,
   that function triggers the napi scheduling.
 * The driver uses CQ for RX for both modes UD and CM, and CQ for TX
   for CM and UD.

Cc: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/ipoib: Get rid of the tx_outstanding variable in all modes

The first step toward using NAPI in the UD/TX flow is to separate
between two flows, the NAPI and the xmit, meaning no use of shared
variables between both flows.

This patch takes out the tx_outstanding variable that was used in both
flows and instead the driver uses the 2 cyclic ring variables: tx_head
and tx_tail, tx_head used in the xmit flow and tx_tail in the NAPI flow.

Cc: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA: Remove Sean's and Hal's emails from MAINTAINER file

RDMA subsystem has one active maintainer who can apply
patches - Doug Ledford, but the RDMA entries in MAINTAINER file
are not stating it. The following patch removes Sean's and Hal's
emails from the maintainers list.

It will allow clean get_maintaner.pl output which is needed for people
outside of our community and various semi-automatic tools.

Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Acked-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Move cqp_cmd_head init to CQP initialization

Control QP (CQP) command backlog list is initialized at
device initialization time.  It is not reinitialized in
the reset flow.  Move the initialization to CQP creation
time so the list can be initialized correctly for reset as well.

Fixes: 86dbcd0f12e9 ("i40iw: add file to handle cqp calls")
Signed-off-by: Bob Sharp <Robert.O.Sharp@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Remove UDA QP from QoS list if creation fails

If User-space Direct Access (UDA) QP creation fails,
the QP entry is not removed from QoS list. Fix this
by removing QP from QoS list if create QP fails.

Fixes: 0fc2dc58896f ("i40iw: Add Quality of Service support")
Signed-off-by: Ivan Barrera <ivan.d.barrera@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Clear CQP Head/Tail during initialization

Clear Control Queue Pair (CQP) Head/Tail on CQP
initialization as during an adapter reset, these
values are not reinitialized. Tail is cleared by
writing 0 to CQP's tail register. Head is cleared
by writing 0 to CQP's doorbell register.

Fixes: 86dbcd0f12e9 ("i40iw: add file to handle cqp calls")
Signed-off-by: Christopher Bednarz <christopher.n.bednarz@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Refactor queue depth calculation

Queue depth calculations use a mix of work requests
and actual number of bytes. Consolidate all calculations
using minimum WQE size to avoid confusion.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Reinitialize IEQ on MTU change

On a netdev MTU change event, the iWARP
Exception Queue (IEQ) buffers may not be
sized properly to handle the new MTU.

Reinitialize the IEQ with new MTU size on MTU
change event.

Also, add define for the max ethernet frame size
field in IEQ QP context instead of the snd_mss
define which is for iWARP QPs' MSS field.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Move ceq_valid to i40iw_sc_dev structure

Completion Event Queues are created and destroyed on
a per device basis as opposed to per User-space Direct
Access resource.

Move ceq_valid to the correct place in i40iw_sc_dev
from i40iw_puda_rsrc.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Account for IPv6 header when setting MSS

The IPv6 header size is not subtracted from MTU when MSS is
set for QPs.

Save MTU opposed to MSS in the vsi struct during
initialization and calculate the MSS based on IPv4 vs
IPv6 connection.

Fixes: f27b4746f378 ("i40iw: add connection management code")
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Remove unused structures

Some structures for post SQ operation are not used.
Remove the following:

i40iw_post_send_w_inv
i40iw_post_inline_send_w_inv
send_w_sol
send_w_inv
send_w_sol_inv
inline_send_w_sol
inline_send_w_inv
inline_send_w_sol_inv

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Move exception_lan_queue to VSI structure

Consolidate exception_lan_queue under VSI structure
where it belongs. Remove it from device and QP structures.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Remove unused static_rsrc from i40iw_create_qp_info

The field static_rsrc in i40iw_create_qp_info is unused
and not needed. Remove it.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Ignore AE source field in AEQE for some AEs

The AE source field in Asynchronous Event Queue Entry (AEQE) is
not set by the hardware for some AEs, but the code assumes it does.
This results in incorrect processing of some AEs.
Fix this by setting ae_src to I40IW_AE_SOURCE_RSVD for those AEs.

Fixes: 86dbcd0f12e9 ("i40iw: add file to handle cqp calls")
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Cleanup AE processing

Remove unimplemented Asynchronous Events (AE) and correct names.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge branch 'timer_setup' into for-next

Conflicts:
drivers/infiniband/hw/cxgb4/cm.c
drivers/infiniband/hw/qib/qib_driver.c
drivers/infiniband/hw/qib/qib_mad.c

There were minor fixups needed in these files.  Just minor context diffs
due to patches from independent sources touching the same basic area.

Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge branch 'for-next-early' into for-next

The early for-next branch was based on v4.14-rc2, while the shared pull
request I got from Mellanox used a v4.14-rc4 base.  I'm making the
branch that was the shared Mellanox pull request the new for-next branch
and merging the early for-next branch into it.

Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Use ARRAY_SIZE

Using the ARRAY_SIZE macro improves the readability of the code.

Found with Coccinelle with the following semantic patch:
@r depends on (org || report)@
type T;
T[] E;
position p;
@@
(
 (sizeof(E)@p /sizeof(*E))
|
 (sizeof(E)@p /sizeof(E[...]))
|
 (sizeof(E)@p /sizeof(T))
)

Signed-off-by: Jérémy Lefaure <jeremy.lefaure@lse.epita.fr>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Fix calculation of maximum RoCE MTU

The original code only took into consideration the largest header
possible after the IB_BTH_BYTES.  This was incorrect, as the largest
possible header size is the largest possible combination of headers we
might run into.  The new code accounts for all possible headers in the
largest possible combination and subtracts that from the MTU to make
sure that all packets will fit on the wire.

Link: https://www.spinics.net/lists/linux-rdma/msg54558.html
Fixes: 3c86aa70bf67 ("RDMA/cm: Add RDMA CM support for IBoE devices")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Reported-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Fix use workqueue without WQ_MEM_RECLAIM

The IB/core provides address resolution service and invokes callback
handler when address resolve request completes of requester in worker
thread context.

Such caller might allocate or free memory in callback handler
depending on the completion status to make further progress or to
terminate a connection. Most ULPs resolve route which involves
allocating route entry and path record elements in callback event handler.

It has been noticed that WQ_MEM_RECLAIM flag should not be used for
workers that tend to allocate memory in this [1] thread discussion.

In order to mitigate this situation, WQ_MEM_RECLAIM flag was dropped for
other such WQs in this [2] patch.

Similar problem might arise with address resolution path, though its not
yet noticed. The ib_addr workqueue is not memory reclaim path due to its
nature of invoking callback that might allocate memory or don't free any
memory under memory pressure.

[1] https://www.spinics.net/lists/linux-rdma/msg53239.html
[2] https://www.spinics.net/lists/linux-rdma/msg53416.html

Fixes: f54816261c2b ("IB/addr: Remove deprecated create_singlethread_workqueue")
Fixes: 5fff41e1f89d ("IB/core: Fix race condition in resolving IP to MAC")
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: Doug Ledford <dledford@redhat.com>

IB/core: Fix unable to change lifespan entry for hw_counters

This patch fixes the case where 'lifespan' entry of the hw_counters
is not writable. Currently write callback is not exposed for for
the hw_counters sysfs operation. Due to this, modifying lifespan
value results into permission denied error in below example.

echo 10 > /sys/class/infiniband/mlx5_0/ports/1/hw_counters/lifespan
-bash: /sys/class/infiniband/mlx5_0/ports/1/hw_counters/lifespan:
Permission denied

This patch adds the hook to modify any attribute which implements
store() operation.

Fixes: b40f4757daa1 ("IB/core: Make device counter infrastructure dynamic")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB: Let ib_core resolve destination mac address

Since IB/core resolves the destination mac address for user and kernel
consumers, avoid resolving in multiple provider drivers.

Only ib_core resolves DMAC now, therefore resolve_eth_dmac is removed as
exported symbol.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Introduce and use rdma_create_user_ah

Introduce rdma_create_user_ah API which allows passing udata to
provider driver and additionally which resolves DMAC for RoCE.

ib_resolve_eth_dmac() resolves destination mac address for unicast,
multicast, link local ipv4 mapped ipv6 and ipv6 destination gid entry.
This allows all RoCE provider drivers to avoid duplicating such code.

Such change brings consistency where IB core always resolves dmac and pass
it to RoCE provider drivers for user and kernel consumers, with this
ah_attr->roce.dmac is always an input field for provider drivers.

This uniformity avoids exporting ib_resolve_eth_dmac symbol to providers
or other modules. Therefore its removed as exported symbol at later in
the patch series.

Now uverbs and umad both makes use of rdma_create_user_ah API which
fixes the issue where umad has invalid DMAC for address.

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: Doug Ledford <dledford@redhat.com>

RDMA/cxgb4: Convert timers to use timer_setup()

In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly. Also removes an unused timer and
drops a redundant initialization.

Cc: Steve Wise <swise@chelsio.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/i40iw: Convert timers to use timer_setup() (part 2)

In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly.

This includes the remaining timers missed in the earlier i40iw patch.

Cc: Faisal Latif <faisal.latif@intel.com>
Cc: Shiraz Saleem <shiraz.saleem@intel.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/cxgb3: Convert timers to use timer_setup()

In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly. Also removes an unused timer.

Cc: Steve Wise <swise@chelsio.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Convert timers to use timer_setup()

In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly. Switches test of .data field to
.function, since .data will be going away.

Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rdmavt: Convert timers to use timer_setup()

In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly.

setup_timer() was already being called before the open-coded init_timer()
and .data assignment. These are removed as well.

Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge tag 'mlx5-updates-2017-10-11' of git://git./linux/kernel/git/mellanox/linux into mlnx-shared

mlx5-updates-2017-10-11: IPoIB Muli Pkey support

This series provides the support for IPoIB Multi Pkey.  InfiniBand Pkeys
are the equivalent of Ethernet vlans.  Currently IPoIB device driver
supports only default Pkey and IPoIB Pkey child interfaces are not
supported with IPoIB offloads mode, this series will add the support for
that by allowing creating mlx5 multiple IPoIB netdevices with a
non-default Pkey.

mlx5 IPoIB Pkey child interface is smaller version of mlx5i IPoIB
interfaces and shares most of its resources with the parent IPoIB
interface, namely RX steering and ring queue resources.

The only mlx5 resources a child Pkey interface will be creating are the
TX rings, since they should be assigned to a specific Pkey.

mlx5i Pkey netdev is implemented via new mlx5e netdev profile
implemented in mlx5/core/ipoib/ipoib_vlan.c.

The series starts with a refactoring of mlx5e PTP and mlx5 clock
implementation to move the code to be part of mlx5 core rather than
mlx5e netdevice, in order to make mlx5 clock and PTP registration part
of the core to be shared with mlx5e master Ethernet netdev/IPoIB parent
netdev and mlx5_ib in the near future.

Add the support for attaching multiple underlay QPs for the different
Pkeys in mlx5 core RX steering.

Add Pkey index to rdma_netdev to add the ability to set PKEY index to
lower IPoIB offload netdev.

Use hash-table to map between DQPN (Destination QP number) to child
netdev for the IPoIB parent netdev to forward RX packets to the
corresponding child Pkey netdev, since the RX rings are shared.

The reset of the series adds the ipoib child Pkey: mlx5e netdev profile,
netdev nods implementation and minimal set of ethtool callbacks.

Thanks,
Saeed.

Signed-off-by: Doug Ledford <dledford@redhat.com>