IB/mlx5: Create profile infrastructure to add and remove stages

Today we have single function which is used when we add an IB interface,
break this function into multiple functions.

Create stages and a generic mechanism to execute each stage.
This is in preparation for RDMA/IB representors which might not need
all stages or will do things differently in some of the stages.

This patch doesn't change any functionality.

Signed-off-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Add RQ/SRQ information to QP stats

When debugging issues with RC QPs, it is useful to know if a QP
has an associated RQ or SRQ, the size of the RQ, and any RNR timeout
values.

Add the necessary information to the QP stats output.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@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: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Assign dest_qp when deregistering mr

It needs to create eight reserve QPs for resolving
a bug of hip06. When deregistering mr, it will issue
a rdma write for every reserve QPs.

When modify qp from init to rtr, it needs to set
the value of dest_qp_num. Otherwise, it will lead
an error of freeing mr.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Fix QP state judgement before sending work requests

The QP can accept send work requests only when the QP is
in the states that allow them to be submitted.

This patch updates the QP state judgement based on the
specification.

Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Filter for zero length of sge in hip08 kernel mode

When the length of sge is zero, the driver need to filter it

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Set access flags of hip08 RoCE

This patch refactors the code of setting access flags
for RDMA operation as well as adds the scene when
attr->max_dest_rd_atomic is zero.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Update the usage of sr_max and rr_max field

This patch fixes the usage with sr_max filed and rr_max of qp
context when modify qp. Its modifications include:
1. Adjust location of filling sr_max filed of qpc
2. Only assign the number of responder resource if
   IB_QP_MAX_DEST_RD_ATOMIC bit is set
3. Only assign the number of outstanding resource if
   IB_QP_MAX_QP_RD_ATOMIC
4. Fix the assgin algorithms for the field of sr_max
   and rr_max of qp context

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Add rq inline data support for hip08 RoCE

This patch mainly implement rq inline data feature for hip08
RoCE in kernel mode.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/ipoib: Fix for notify send CQ failure messages

If IB_CQ_REPORT_MISSED_EVENTS flag is passed in ib_req_notify_cq()
it may return positive value indicating non-empty CQ.
If return code not verified the log might be flooded with false
warning messages "request notify on send CQ failed".

Fixes: 8966e28d2e40 ("IB/ipoib: Use NAPI in UD/TX flows")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/cma: Mark end of CMA ID messages

The commit 1a1c116f3dcf ("RDMA/netlink: Simplify the put_msg and put_attr")
removes nlmsg_len calculation in ibnl_put_attr causing netlink messages and
caused to miss source and destination addresses.

Fixes: 1a1c116f3dcf ("RDMA/netlink: Simplify the put_msg and put_attr")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/nldev: Refactor setting the nldev handle to a common function

The NLDEV commands are using IB device indexes and names as a handle
for netlink communications. Put all relevant code into one function
to remove code duplication in followup patches.

Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Replace open-coded variant of put_device

There is an existing function to decrease reference counter
of the device, let's use it.

Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/netlink: Simplify code of autoload modules

The request_module() call is internally wrapped by CONFIG_MODULE,
so there is no need to check it in our RDMA code too.

Refactor to simplify the code.

Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/rxe: Remove useless EXPORT_SYMBOL

The RXE driver is standalone module and hence doesn't need to export
symbols, nor does this one line function deserve to be not inlined.

Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mthca: Use zeroing memory allocator than allocator/memset

Use dma_zalloc_coherent for allocating zeroed
memory and remove unnecessary memset function.

Done using Coccinelle.
Generated-by: scripts/coccinelle/api/alloc/kzalloc-simple.cocci
0-day tested with no failures.

Suggested-by: Luis R. Rodriguez <mcgrof@kernel.org>
Signed-off-by: Himanshu Jha <himanshujha199640@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/bnxt_re: Use zeroing memory allocator than allocator/memset

Use dma_zalloc_coherent for allocating zeroed
memory and remove unnecessary memset function.

Done using Coccinelle.
Generated-by: scripts/coccinelle/api/alloc/kzalloc-simple.cocci
0-day tested with no failures.

Suggested-by: Luis R. Rodriguez <mcgrof@kernel.org>
Signed-off-by: Himanshu Jha <himanshujha199640@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/qedr: Use zeroing memory allocator than allocator/memset

Use dma_zalloc_coherent for allocating zeroed
memory and remove unnecessary memset function.

Done using Coccinelle.
Generated-by: scripts/coccinelle/api/alloc/kzalloc-simple.cocci
0-day tested with no failures.

Suggested-by: Luis R. Rodriguez <mcgrof@kernel.org>
Signed-off-by: Himanshu Jha <himanshujha199640@gmail.com>
Acked-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Use zeroing memory allocator than allocator/memset

Use dma_zalloc_coherent for allocating zeroed
memory and remove unnecessary memset function.

Done using Coccinelle.
Generated-by: scripts/coccinelle/api/alloc/kzalloc-simple.cocci
0-day tested with no failures.

Suggested-by: Luis R. Rodriguez <mcgrof@kernel.org>
Signed-off-by: Himanshu Jha <himanshujha199640@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/ocrdma: Use zeroing memory allocator than allocator/memset

Use dma_zalloc_coherent for allocating zeroed
memory and remove unnecessary memset function.

Done using Coccinelle.
Generated-by: scripts/coccinelle/api/alloc/kzalloc-simple.cocci
0-day tested with no failures.

Suggested-by: Luis R. Rodriguez <mcgrof@kernel.org>
Signed-off-by: Himanshu Jha <himanshujha199640@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

iw_cxgb4: Change error/warn prints to pr_debug

These prints not neccesarily mean error/warning, so changing them to
debug.

Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Add detailed comments for mb() call

This patch adds more detailed comments when we call the
memory barrier function, such as rmb, wmb and mb. Three
mb() callers are deleted since they are unnecessary.

Suggested-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Enable QP creation with a given blue flame index

This patch enables QP creation with a given BF index, this allows the
user space driver to share same BF between few QPs or alternatively have
a dedicated BF per QP.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Expose dynamic mmap allocation

This patch exposes the option to dynamic allocates a UAR, this
functionality will be used in downstream patch in this series as
part of QP creation.

Specifically, the user space driver asks for a UAR allocation in a given
page index, upon success this UAR and its bfregs can be used as part of
QP creation by the user space driver.

To enable allocating more than 256 UARs the page index is encoded in an
extra one byte just after the command byte.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Extend UAR stuff to support dynamic allocation

This patch extends the alloc context flow to be prepared for working
with dynamic UAR allocations.

Currently upon alloc context there is some fix size of UARs that are
allocated (named 'static allocation') and there is no option to user
application to ask for more or control which UAR will be used by which
QP.

In this patch the driver prepares its data structures to manage both the
static and the dynamic allocations and let the user driver knows about
the max value of dynamic blue-flame registers that are allowed.

Downstream patches from this series will enable the dynamic allocation
and the association as part of QP creation.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Report inner RSS capability

Add missing inner RSS support capability as part of
the RSS supported fields.

In addition change MLX5_RX_HASH_INNER to 1UL << 31 in
order to define it as unsigned.

Fixes: 309fa3470fca ("IB/mlx5: Add support for RSS on the inner packet")
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx4: Add support to RSS hash for inner headers

Support RSS hash for inner headers according to a new flag,
MLX4_IB_RX_HASH_INNER provided by the vendor channel.

In case the flag is set, RSS hash will be done on the inner headers of
VXLAN packets (which are encapsulated).
Non-encapsulated packets will be hashed according to the outer headers.

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: Jason Gunthorpe <jgg@mellanox.com>

Merge branch 'from-rc' of git://git./linux/kernel/git/rdma/rdma.git

Patches for 4.16 that are dependent on patches sent to 4.15-rc.

These are small clean ups for the vmw_pvrdma and i40iw drivers.

* 'from-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git:
  RDMA/vmw_pvrdma: Remove usage of BIT() from UAPI header
  RDMA/vmw_pvrdma: Use refcount_t instead of atomic_t
  RDMA/vmw_pvrdma: Use more specific sizeof in kcalloc
  RDMA/vmw_pvrdma: Clarify QP and CQ is_kernel logic
  RDMA/vmw_pvrdma: Add UAR SRQ macros in ABI header file
  i40iw: Change accelerated flag to bool

RDMA/vmw_pvrdma: Remove usage of BIT() from UAPI header

BIT() should not be used in the UAPI header. Remove it.

Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Use refcount_t instead of atomic_t

refcount_t is the preferred type for refcounts. Change the
QP and CQ refcnt fields to use refcount_t.

Reviewed-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Use more specific sizeof in kcalloc

Convert the sizeof(void *) in two kcalloc calls to be more
specific for the arrays that are being allocated.

Reviewed-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Clarify QP and CQ is_kernel logic

Be more consistent in setting and checking is_kernel
flag for QPs and CQs.

Reviewed-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Add UAR SRQ macros in ABI header file

Support for SRQs were added in the vmw_pvrdma userlevel library
before two necessary macros were added into the kernel ABI header
file. Add the two UAR SRQ macros that are required by the userlevel
library so that the library can rely on the kernel ABI header file
for these SRQ macro definitions.

Fixes: 8b10ba783c9d ("RDMA/vmw_pvrdma: Add shared receive queue support")
Reviewed-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Change accelerated flag to bool

The accelerated flag only utilizes two values: 0 and 1.
Modify accelerated flag in struct i40iw_cm_node to bool.

Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

infiniband: drop unknown function from core_priv.h

Delete ibnl_chk_listeners() and its kernel-doc comments from the
core_priv.h header file.  There is no such function.

Fixes: 233c1955835b ("RDMA/netlink: Reduce exposure of RDMA netlink functions")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/core: Make sure that PSN does not overflow

The rq/sq->psn is 24 bits as defined in the IB spec, therefore we mask
out the 8 most significant bits to avoid overflow in modify_qp.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Fix mlx5_ib_alloc_mr error flow

ibmr.device is being set only after ib_alloc_mr() is
(successfully) complete. Therefore, in case mlx5_core_create_mkey()
return with error, the error flow calls mlx5_free_priv_descs()
which uses ibmr.device (which doesn't exist yet), causing
a NULL dereference oops.

To fix this, the IB device should be set in the mr struct earlier
stage (e.g. prior to calling mlx5_core_create_mkey()).

Fixes: 8a187ee52b04 ("IB/mlx5: Support the new memory registration API")
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Nitzan Carmi <nitzanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/core: Verify that QP is security enabled in create and destroy

The XRC target QP create flow sets up qp_sec only if there is an IB link with
LSM security enabled. However, several other related uAPI entry points blindly
follow the qp_sec NULL pointer, resulting in a possible oops.

Check for NULL before using qp_sec.

Cc: <stable@vger.kernel.org> # v4.12
Fixes: d291f1a65232 ("IB/core: Enforce PKey security on QPs")
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/uverbs: Fix command checking as part of ib_uverbs_ex_modify_qp()

If the input command length is larger than the kernel supports an error should
be returned in case the unsupported bytes are not cleared, instead of the
other way aroudn. This matches what all other callers of ib_is_udata_cleared
do and will avoid user ABI problems in the future.

Cc: <stable@vger.kernel.org> # v4.10
Fixes: 189aba99e700 ("IB/uverbs: Extend modify_qp and support packet pacing")
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Serialize access to the VMA list

User-space applications can do mmap and munmap directly at
any time.

Since the VMA list is not protected with a mutex, concurrent
accesses to the VMA list from the mmap and munmap can cause
data corruption. Add a mutex around the list.

Cc: <stable@vger.kernel.org> # v4.7
Fixes: 7c2344c3bbf9 ("IB/mlx5: Implements disassociate_ucontext API")
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Change slid arg in ingress_pkey_table_fail to 32bit

Change the slid arg to ingress_pkey_table_fail() to a full
32Bits and do not convert to 16Bits in caller. This is so we
can keep everything 32bit in the kernel and only change to
16bit at the uapi boundary.

Signed-off-by: Don Hiatt <don.hiatt@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/core: Use rdma_cap_opa_mad to check for OPA

Use rdma_cap_opa_mad() to check for OPA to promote code reuse.

Signed-off-by: Don Hiatt <don.hiatt@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Fix the connection ORD value for loopback

The accepting QP ORD value should be adjusted not to
exceed the peer QP IRD value (RFC 6581). This is
skipped for loopback. After the ORD is validated
by i40iw_record_ird_ord(), adjust the ORD value of
the loopback accepting QP to prevent overrunning the
IRD space of the peer QP. Also move the ORD accounting
for 0-byte RDMA read to i40iw_record_ird_ord().

Fixes: f27b4746f378 ("i40iw: add connection management code")
Signed-off-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Validate correct IRD/ORD connection parameters

Casting to u16 before validating IRD/ORD connection
parameters could cause recording wrong IRD/ORD values
in the cm_node. Validate the IRD/ORD parameters as
they are passed by the application before recording
them.

Fixes: f27b4746f378 ("i40iw: add connection management code")
Signed-off-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Ignore LLP_DOUBT_REACHABILITY AE

The LLP_DOUBT_REACHABILITY Asynchronous Event (AE) is an early
warning of a connection issue. It is followed by LLP_TOO_MANY_RETRIES
AE, if the retransmit threshold is reached and recovery is not possible
for the connection.

Currently we terminate the connection on receiving the
LLP_DOUBT_REACHABILITY AE. Ignore this AE and
terminate the connection only on LLP_TOO_MANY_RETRIES AE.

This improves the user experience on cable disconnect/reconnect
scenario while running iWARP traffic. On cable disconnect,
the QP traffic is paused and the user has a larger and more
reasonable timeout within which if the cable is reconnected,
traffic can continue.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Fix sequence number for the first partial FPDU

Partial FPDU processing is broken as the sequence number
for the first partial FPDU is wrong due to incorrect
Q2 buffer offset. The offset should be 64 rather than 16.

Fixes: 786c6adb3a94 ("i40iw: add puda code")
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Selectively teardown QPs on IP addr change event

On IP address change event, all connected QPs are torn down
irrespective of whether IP address is involved in a connection.

Only teardown connections those source or destination address
matches the netdev interface IP address being changed, and if
they are on the same VLAN as the netdev.

Fixes: e5e74b61b165 ("i40iw: Add IP addr handling on netdev events")
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Add notifier for network device events

Register a netdevice notifier for netdev UP/DOWN
notification events and report the appropriate ib event.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Correct Q1/XF object count equation

Lower Inbound RDMA Read Queue (Q1) object count by a factor of 2
as it is incorrectly doubled. Also, round up Q1 and Transmit FIFO (XF)
object count to power of 2 to satisfy hardware requirement.

Fixes: 86dbcd0f12e9 ("i40iw: add file to handle cqp calls")
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Use utility function roundup_pow_of_two()

Consolidate all power of 2 round calculations to
use kernel utility function roundup_pow_of_two().

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

i40iw: Set MAX_IRD_SIZE to 64

Increase I40IW_MAX_IRD_SIZE to 64 which is the device limit.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

rdma: Update maintainer contact for Intel RDMA drivers

Ensure both Mike and I are listed as maintainer contacts for Intel's qib,
hfi1, and rdmavt drivers.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/SA: Check dlid before SA agent queries for ClassPortInfo

SA queries SM for class port info when there is a LID_CHANGE event.

When a base lid is configured before fm is started ie when smlid is
not yet assigned, SA handles the LID_CHANGE event and tries query SM
with lid 0. This will cause an hang.

[ 1106.958820] INFO: task kworker/2:0:23 blocked for more than 120 seconds.
[ 1106.965082] Tainted: G O 4.12.0+ #1
[ 1106.969602] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables
 this message.
[ 1106.977227] kworker/2:0 D 0 23 2 0x00000000
[ 1106.977250] Workqueue: infiniband update_ib_cpi [ib_core]
[ 1106.977261] Call Trace:
[ 1106.977273] __schedule+0x28e/0x860
[ 1106.977285] schedule+0x36/0x80
[ 1106.977298] schedule_timeout+0x1a3/0x2e0
[ 1106.977310] ? radix_tree_iter_tag_clear+0x1b/0x20
[ 1106.977322] ? idr_alloc+0x64/0x90
[ 1106.977334] wait_for_completion+0xe3/0x140
[ 1106.977347] ? wake_up_q+0x80/0x80
[ 1106.977369] update_ib_cpi+0x163/0x210 [ib_core]
[ 1106.977381] process_one_work+0x147/0x370
[ 1106.977394] worker_thread+0x4a/0x390
[ 1106.977406] kthread+0x109/0x140
[ 1106.977418] ? process_one_work+0x370/0x370
[ 1106.977430] ? kthread_park+0x60/0x60
[ 1106.977443] ret_from_fork+0x22/0x30

Always ensure a proper smlid is assigned before querying SM for cpi.

Fixes: ee1c60b1bff ("IB/SA: Modify SA to implicitly cache Class Port info")
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Venkata Sandeep Dhanalakota <venkata.s.dhanalakota@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

nes: Change accelerated flag to bool

The accelerated flag only utilizes two values: 0 and 1.
Modify accelerated flag in struct nes_cm_node to bool.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi: Only read capability registers if the capability exists

During driver init, various registers are saved to allow restoration
after an FLR or gen3 bump.  Some of these registers are not available
in some circumstances (i.e. Virtual machines).

This bug makes the driver unusable when the PCI device is passed into
a VM, it fails during probe.

Delete unnecessary register read/write, and only access register if
the capability exists.

Cc: <stable@vger.kernel.org> # 4.14.x
Fixes: a618b7e40af2 ("IB/hfi1: Move saving PCI values to a separate function")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@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: Jason Gunthorpe <jgg@mellanox.com>

drivers: infiniband: remove duplicate includes

These duplicate includes have been found with scripts/checkincludes.pl but
they have been removed manually to avoid removing false positives.

Signed-off-by: Pravin Shedge <pravin.shedge4linux@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Add eq support of hip08

This patch adds eq support for hip08. The eq table can
be multi-hop addressed.

Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Reviewed-by: Lijun Ou <oulijun@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Refactor eq code for hip06

Considering the compatibility of supporting hip08's eq
process and possible changes of data structure, this patch
refactors the eq code structure of hip06.

We move all the eq process code for hip06 from hns_roce_eq.c
into hns_roce_hw_v1.c, and also for hns_roce_eq.h. With
these changes, it will be convenient to add the eq support
for later hardware version.

Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Reviewed-by: Lijun Ou <oulijun@huawei.com>
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/ipoib: Fix lockdep issue found on ipoib_ib_dev_heavy_flush

The locking order of vlan_rwsem (LOCK A) and then rtnl (LOCK B),
contradicts other flows such as ipoib_open possibly causing a deadlock.
To prevent this deadlock heavy flush is called with RTNL locked and
only then tries to acquire vlan_rwsem.
This deadlock is possible only when there are child interfaces.

[  140.941758] ======================================================
[  140.946276] WARNING: possible circular locking dependency detected
[  140.950950] 4.15.0-rc1+ #9 Tainted: G           O
[  140.954797] ------------------------------------------------------
[  140.959424] kworker/u32:1/146 is trying to acquire lock:
[  140.963450]  (rtnl_mutex){+.+.}, at: [<ffffffffc083516a>] __ipoib_ib_dev_flush+0x2da/0x4e0 [ib_ipoib]
[  140.970006]
but task is already holding lock:
[  140.975141]  (&priv->vlan_rwsem){++++}, at: [<ffffffffc0834ee1>] __ipoib_ib_dev_flush+0x51/0x4e0 [ib_ipoib]
[  140.982105]
which lock already depends on the new lock.
[  140.990023]
the existing dependency chain (in reverse order) is:
[  140.998650]
-> #1 (&priv->vlan_rwsem){++++}:
[  141.005276]        down_read+0x4d/0xb0
[  141.009560]        ipoib_open+0xad/0x120 [ib_ipoib]
[  141.014400]        __dev_open+0xcb/0x140
[  141.017919]        __dev_change_flags+0x1a4/0x1e0
[  141.022133]        dev_change_flags+0x23/0x60
[  141.025695]        devinet_ioctl+0x704/0x7d0
[  141.029156]        sock_do_ioctl+0x20/0x50
[  141.032526]        sock_ioctl+0x221/0x300
[  141.036079]        do_vfs_ioctl+0xa6/0x6d0
[  141.039656]        SyS_ioctl+0x74/0x80
[  141.042811]        entry_SYSCALL_64_fastpath+0x1f/0x96
[  141.046891]
-> #0 (rtnl_mutex){+.+.}:
[  141.051701]        lock_acquire+0xd4/0x220
[  141.055212]        __mutex_lock+0x88/0x970
[  141.058631]        __ipoib_ib_dev_flush+0x2da/0x4e0 [ib_ipoib]
[  141.063160]        __ipoib_ib_dev_flush+0x71/0x4e0 [ib_ipoib]
[  141.067648]        process_one_work+0x1f5/0x610
[  141.071429]        worker_thread+0x4a/0x3f0
[  141.074890]        kthread+0x141/0x180
[  141.078085]        ret_from_fork+0x24/0x30
[  141.081559]

other info that might help us debug this:
[  141.088967]  Possible unsafe locking scenario:
[  141.094280]        CPU0                    CPU1
[  141.097953]        ----                    ----
[  141.101640]   lock(&priv->vlan_rwsem);
[  141.104771]                                lock(rtnl_mutex);
[  141.109207]                                lock(&priv->vlan_rwsem);
[  141.114032]   lock(rtnl_mutex);
[  141.116800]
 *** DEADLOCK ***

Fixes: b4b678b06f6e ("IB/ipoib: Grab rtnl lock on heavy flush when calling ndo_open/stop")
Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Fix congestion counters in LAG mode

Congestion counters are counted and queried per physical function.
When working in LAG mode, CNP packets can be sent or received on both
of the functions, thus congestion counters should be aggregated from
the two physical functions.

Fixes: e1f24a79f424 ("IB/mlx5: Support congestion related counters")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Aviv Heller <avivh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Avoid use after free due to QP/CQ/SRQ destroy

The use of wait queues in vmw_pvrdma for handling concurrent
access to a resource leaves a race condition which can cause a use
after free bug.

Fix this by using the pattern from other drivers, complete() protected by
dec_and_test to ensure complete() is called only once.

Fixes: 29c8d9eba550 ("IB: Add vmw_pvrdma driver")
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Use refcount_dec_and_test to avoid warning

refcount_dec generates a warning when the operation
causes the refcount to hit zero. Avoid this by using
refcount_dec_and_test.

Fixes: 8b10ba783c9d ("RDMA/vmw_pvrdma: Add shared receive queue support")
Reviewed-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/vmw_pvrdma: Call ib_umem_release on destroy QP path

The QP cleanup did not previously call ib_umem_release,
resulting in a user-triggerable kernel resource leak.

Fixes: 29c8d9eba550 ("IB: Add vmw_pvrdma driver")
Reviewed-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

iw_cxgb4: when flushing, complete all wrs in a chain

If a wr chain was posted and needed to be flushed, only the first
wr in the chain was completed with FLUSHED status.  The rest were
never completed.  This caused isert to hang on shutdown due to the
missing completions which left iscsi IO commands referenced, stalling
the shutdown.

Fixes: 4fe7c2962e11 ("iw_cxgb4: refactor sq/rq drain logic")

Cc: stable@vger.kernel.org
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

iw_cxgb4: reflect the original WR opcode in drain cqes

The flush/drain logic was not retaining the original wr opcode in
its completion.  This can cause problems if the application uses
the completion opcode to make decisions.

Use bit 10 of the CQE header word to indicate the CQE is a special
drain completion, and save the original WR opcode in the cqe header
opcode field.

Fixes: 4fe7c2962e11 ("iw_cxgb4: refactor sq/rq drain logic")

Cc: stable@vger.kernel.org
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

iw_cxgb4: Only validate the MSN for successful completions

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>

RDMA/ocrdma: Fix permissions for OCRDMA_RESET_STATS

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>

iser-target: avoid reinitializing rdma contexts for isert commands

isert commands that failed during isert_rdma_rw_ctx_post() are queued to
Queue-Full(QF) queue and are scheduled to be reposted during queue-full
queue processing. During this reposting, the rdma contexts are initialised
again in isert_rdma_rw_ctx_post(), which is leaking significant memory.

unreferenced object 0xffff8830201d9640 (size 64):
  comm "kworker/0:2", pid 195, jiffies 4295374851 (age 4528.436s)
  hex dump (first 32 bytes):
    00 60 8b cb 2e 00 00 00 00 10 00 00 00 00 00 00  .`..............
    00 90 e3 cb 2e 00 00 00 00 10 00 00 00 00 00 00  ................
  backtrace:
    [<ffffffff8170711e>] kmemleak_alloc+0x4e/0xb0
    [<ffffffff811f8ba5>] __kmalloc+0x125/0x2b0
    [<ffffffffa046b24f>] rdma_rw_ctx_init+0x15f/0x6f0 [ib_core]
    [<ffffffffa07ab644>] isert_rdma_rw_ctx_post+0xc4/0x3c0 [ib_isert]
    [<ffffffffa07ad972>] isert_put_datain+0x112/0x1c0 [ib_isert]
    [<ffffffffa07dddce>] lio_queue_data_in+0x2e/0x30 [iscsi_target_mod]
    [<ffffffffa076c322>] target_qf_do_work+0x2b2/0x4b0 [target_core_mod]
    [<ffffffff81080c3b>] process_one_work+0x1db/0x5d0
    [<ffffffff8108107d>] worker_thread+0x4d/0x3e0
    [<ffffffff81088667>] kthread+0x117/0x150
    [<ffffffff81713fa7>] ret_from_fork+0x27/0x40
    [<ffffffffffffffff>] 0xffffffffffffffff

Here is patch to use the older rdma contexts while reposting
the isert commands intead of reinitialising them.

Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/cm: Refactor to avoid setting path record software only fields

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>

IB/{core, umad, cm}: Rename ib_init_ah_from_wc to ib_init_ah_attr_from_wc

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>

IB/{core, cm, cma, ipoib}: Rename ib_init_ah_from_path to ib_init_ah_attr_from_path

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>

IB/cm: Fix sleeping while spin lock is held

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>

IB/cm: Handle address handle attribute init error

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>

IB/{cm, umad}: Handle av init error

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>

IB/{core, ipoib}: Simplify ib_find_gid to search only for IB link layer

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/core: Avoid copying ifindex twice

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>

RDMA/{core, cma}: Simplify rdma_translate_ip

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>

IB/core: Removed unused function

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/core: Avoid redundant memcpy in rdma_addr_find_l2_eth_by_grh

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/core: Avoid exporting module internal ib_find_gid_by_filter()

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>

IB/rxe: Avoid passing unused index pointer which is optional

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>

IB/core: Refactor to avoid unnecessary check on GID lookup miss

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>

IB/core: Avoid unnecessary type cast

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>

RDMA/cma: Introduce and use helper functions to init work

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>

RDMA/cma: Avoid setting path record type twice

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>

RDMA/cma: Simplify netdev check

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>

RDMA/cma: Set default GID type as RoCE when resolving RoCE route

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>

IB/umem: Fix use of npages/nmap fields

In ib_umem structure npages holds original number of sg entries, while
nmap is number of DMA blocks returned by dma_map_sg.

Fixes: c5d76f130b28 ('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>

IB/cm: Add debug prints to ib_cm

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>

IB/core: Fix memory leak in cm_req_handler error flows

In cm_req_handler error flows, sometimes cm_id_priv->timewait_info
isn't free'd.

Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Mukesh Kacker <mukesh.kacker@oracle.com>
Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/cma: Use correct size when writing netlink stats

The code was using the src size when formatting the dst. They are almost
certainly the same value but it reads wrong.

Fixes: ce117ffac2e9 ("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>

IB/ipoib: Update pathrec field if not valid record

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>

IB/ipoib: Avoid memory leak if the SA returns a different DGID

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/mlx4: Remove unused ibpd parameter

Remove unused ibpd parameter from create_qp_rss() function.

Signed-off-by: Erez Alfasi <ereza@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/core: Avoid exporting module internal function

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>

IB/core: Depend on IPv6 stack to resolve link local address for RoCEv2

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>

IB/{core/cm}: Fix generating a return AH for RoCEE

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>

IB/ocrdma: Remove unneeded conversions to bool

Found with scripts/coccinelle/misc/boolconv.cocci.

Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx4: Potential buffer overflow in _mlx4_set_path()

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>

RDMA/cxgb4: Add a sanity check in process_work()

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>

iw_cxgb4: make pointer reg_workq static

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: 1c8f1da5d851 ("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>