IB/rdmavt: Trivial function comment corrected.

Corrected function name in comment from qib_ to rvt_.

Signed-off-by: Parav Pandit <pandit.parav@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix trace of atomic ack

The length is incorrect, causing the trace data to
be truncated.

Add the additional 8 bytes that should have been there.
Also trace out the atomic ack in hex to aid debugging.

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: Update SMA ingress checks for response packets

Fix "unsupported method" error by skipping ingress pkey checks on
response SMA packets.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Use EPROM platform configuration read

The driver will now try to read directly from the EPROM as its
first choice for the platform configuration file.

Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add ability to read platform config from the EPROM

Add a function to read the platform configuration file from
the EPROM.

Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Restore EPROM read ability

Partially revert commit d07903174202 ("IB/hfi1: Remove
EPROM functionality from data device"), bringing back
the ability to read from the EPROM.

This code will be used for driver-only acccess to the EPROM, hence
change EPROM read to save to a buffer instead of copy touser. Also
allow any offset and remove missed includes and leftover declarations.

Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Document new sysfs entries for hfi1 driver

This patch adds description for the sdma engine related sysfs entries
for the HFI1 driver.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add new debugfs sdma_cpu_list file

Add a debugfs sdma_cpu_list file that can be used to examine the CPU to
sdma engine assignments for the whole device.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add irq affinity notification handler

This patch adds an irq affinity notification handler.
When a user changes interrupt affinity settings for an sdma engine,
the driver needs to make changes to its internal sde structures and
also update the affinity_hint.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add a new VL sysfs attribute for sdma engines

This patch adds a read-only "VL" attribute for the sysfs entry of each
sdma engine. It will allow the user to check VL to sdma engine mappings.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add sysfs interface for affinity setup

Some users want more control over which cpu cores are being used by the
driver. For example, users might want to restrict the driver to some
specified subset of the cores so that they can appropriately partition
processes, irq handlers, and work threads.
To allow the user to fine tune system affinity settings new sysfs
attributes are introduced per sdma engine.  This patch adds a new
attribute type for sdma engine and a new cpu_list attribute.
When the user writes a cpu range to the cpu_list attribute the driver
will create an internal cpu->sdma map, which will be used later as a
look-up table to choose an optimal engine for a user requests.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix resource release in context allocation

Correct resource free in allocate_ctxt() function.
When context creation fails allocated resources are properly
released and pointer in receive context data table is set back
to NULL.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jakub Pawlak <jakub.pawlak@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Remove unused variable from devdata

We no longer use an error tasklet. Remove it from the hfi1_devdata
structure.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Cleanup tasklet refs in comments

The code no longer uses tasklets for the send engine. However it does
use a tasklet for sdma but the send routines use a workqueue now days.
Update the comments to reflect that. Make things more generic with
saying "send engine" because that is what is being referred to.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Adjust hardware buffering parameter

It was determined that 0x880 is a better value for hardware buffering,
use it.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Act on external device timeout

Add missing external device timeout notification.  Recognize
it as a failed LNI signal from the 8051 firmware.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix defered ack race with qp destroy

There is a a bug in defered ack stuff that causes a race with the
destroy of a QP.

A packet causes a defered ack to be pended by putting the QP
into an rcd queue.

A return from the driver interrupt processing will process that rcd
queue of QPs and attempt to do a direct send of the ack.   At this
point no locks are held and the above QP could now be put in the reset
state in the qp destroy logic.   A refcount protects the QP while it
is in the rcd queue so it isn't going anywhere yet.

If the direct send fails to allocate a pio buffer,
hfi1_schedule_send() is called to trigger sending an ack from the
send engine. There is no state test in that code path.

The refcount is then dropped from the driver.c caller
potentially allowing the qp destroy to continue from its
refcount wait in parallel with the workqueue scheduling of the qp.

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

IB/hfi1: Combine shift copy and byte copy for SGE reads

Prevent over-reading the SGE length by using byte
reads for non quad-word reads.

Reviewed-by: Dean Luick <dean.luick@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: Do not read more than a SGE length

In certain cases, if the tail of an SGE is not
8-byte aligned, bytes beyond the end to an 8-byte
alignment can be read. Change the copy routine
to avoid the over-read. Instead, stop on the final
whole quad-word, then read the remaining bytes.

Reviewed-by: Dean Luick <dean.luick@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: Extend i2c timeout

Allow a longer timeout for i2c due to clock stretching and
inaccurate jiffy timing when under a spin lock.  This timeout
is consistent with other i2c-algo-bit users.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Increase default settings of max_cqes and max_qps

The ib_write_bw test allows using up to 16384 QPs. When a relatively
large number of QPs (within that range) is used, the test can fail
because the number of CQ entries needed exceeds the limit set by the
driver.

This patch increases the default setting of max_cqes from 0x2FFFF
(196607) to 0x2FFFFF(3145727), which is sufficient to cover the
maximum number needed by the ib_write_bw test (2097152). The default
setting of max_qps is also increased from 16384 to 32768 to allow
the test to run successfully with 16383 or 16384 QPs.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Remove filtering of Set(PkeyTable) in HFI SMA

The FM should have full control to set the pkeys in the
driver pkey table. Remove filtering done by the driver.

Reviewed-by: Ira Weiny <ira.weiny@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/qib: Remove qpt_mask global

There is no need to have a global qpt_mask as that does not support the
multiple chip model which qib has. Instead rely on the value which
exists already in the device data (dd).

Fixes: 898fa52b4ac3 "IB/qib: Remove qpn, qp tables and related variables from qib"
Reviewed-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: Consolidate pio control masks into single definition

This allows for adding additional pages of adaptive pio
opcode control including manufacturer specific ones.

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/rdmavt, IB/hfi1: Add lockdep asserts for lock debug

This patch adds lockdep asserts in key code paths for
insuring lock correctness.

Reviewed-by: Ira Weiny <ira.weiny@intel.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/rdmavt: Add qp init function

Add an rvt_qp_init() to initialize specific
common fields as the qp is created or reset.

The routine is shared by the rvt_reset_qp() and
the rvt_create_qp().

The intent is that lock dep assertions will only
appear in the rvt_reset_qp().

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/rdmavt: Move reset calldown to reset path

The reset calldown is misplaced.

It should only be called in the code that actually
transitions the QP to reset.

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: Move iowait_init() to priv allocate

The call is misplaced in the reset calldown function
and causes issues with lockdep assertions that are to
be added.

Fixes: Commit a2c2d608957c ("staging/rdma/hfi1: Remove create_qp functionality")
Reviewed-by: Ira Weiny <ira.weiny@intel.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/rdmavt: Correct sparse annotation

The __must_hold() is sufficent to correct the sparse
context imbalance inside a function.

Per Documentation/sparse.txt:
__must_hold - The specified lock is held on function entry and exit.

Fixes: Commit c0a67f6ba356 ("IB/rdmavt: Annotate rvt_reset_qp()")
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: Fix locking scheme for affinity settings

Existing locking scheme in affinity.c file using the
&node_affinity.lock spinlock is not very elegant.
We acquire the lock to get hfi1_affinity_node entry,
unlock, and then use the entry without the lock held.
With more functions being added, which access and
modify the entries, this can lead to race conditions.
This patch makes this locking scheme more consistent.
It changes the spinlock to mutex. Since all the code
is executed in a user process context there is no need
for a spinlock. This also allows to keep the lock
not only while we look up for the node affinity entry,
but over the whole section where the entry is being used.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix user-space buffers mapping with IOMMU enabled

The dma_XXX API functions return bus addresses which are
physical addresses when IOMMU is disabled. Buffer
mapping to user-space is done via remap_pfn_range() with PFN
based on bus address instead of physical. This results in
wrong pages being mapped to user-space when IOMMU is enabled.

Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Tymoteusz Kielan <tymoteusz.kielan@intel.com>
Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix the count of user packets submitted to an SDMA engine

Each user SDMA request coming into the driver may contain multiple packets.
Each user packet may use multiple SDMA descriptors to fill the send buffer.
The field seqsubmitted in struct user_sdma_request counts the number of
user packets submitted to an SDMA engine. Sometimes, the intermediate count
may not be updated properly. However, once all the packets' descriptors
are successfully submitted to the SDMA engine, the final count is updated
correctly. But, if only some of the packets are submitted to the engine due
to an error, the intermediate count doesn't reflect the partial number of
packets submitted to the SDMA engine. This can cause a hang later in the
code as the count of packets submitted to the SDMA engine doesn't match the
the count of packets processed by the SDMA engine.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Move serdes tune inside link start function

All calls to tune_serdes and start_link are paired.  Move
tune_serdes inside start_link.

Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/qib,IB/hfi: Use core common header file

Use common header file structs, defines, and accessors
in the drivers.  The old declarations are removed.

The repositioning of the includes allows for the removal
of hfi1_message_header and replaces its use with ib_header.

Also corrected are two issues with set_armed_to_active():
- The "packet" parameter is now a pointer as it should have been
- The etype is validated to insure that the header is correct

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Don Hiatt <don.hiatt@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/core: Add ib headers for general use

Add IB headers, defines, and accessors that are identical
in both qib and hfi1 into the core includes.

The accessors for be maintenance of __be64 fields since
alignment is potentially invalid and can differ based on
the presense of the GRH.

{hfi1,qib}_ib_headers will be ib_headers.
{hfi1,qib|_other_headers will be ib_other_headers.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Don Hiatt <don.hiatt@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/rdmavt, IB/qib, IB/hfi1: Use new QP put get routines

This improves readability and hides the reference count
mechanism from the client drivers.

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/rdmavt: Add functions to get and release QP references

This centralizes the function and improves code readability.

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/rdmavt: Don't vfree a kzalloc'ed memory region

The userspace memory region 'mr' is allocated with kzalloc in
__rvt_alloc_mr  however it is incorrectly being freed with vfree in
__rvt_free_mr. Fix this by using kfree to free it.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Fix kmem_cache leak

Decrement qp reference when handling error path
in completer to prevent kmem_cache leak.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Fix race condition between requester and completer

rxe_requester() is sending a pkt with rxe_xmit_packet() and
then calls rxe_update() to update the wqe and qp's psn values.
But sometimes the response is received before the requester
had time to update the wqe in which case the completer
acts on errornous wqe values.
This fix updates the wqe and qp before actually sending
the request and rolls back when xmit fails.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Fix duplicate atomic request handling

When handling ack for atomic opcodes like "fetch&add"
or "cmp&swp", the method send_atomic_ack() saves the ack
before sending it, in case it gets lost and never reach the
requester. In which case the method duplicate_request()
will need to find it using the duplicated request.psn.
But send_atomic_ack() used a wrong psn value and thus
the above ack was never found.
This fix uses the ack.psn to locate the ack in case
its needed.
This fix also copies the ack packet to the skb's control buffer
since duplicate_request() will need it when calling rxe_xmit_packet()

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Fix kernel panic in udp_setup_tunnel

Disable creation of a UDP socket for ipv6 when
CONFIG_IPV6 is not enabeld. Since udp_sock_create6()
returns 0 when CONFIG_IPV6 is not set

[   46.888632] IP: [<c220705a>] setup_udp_tunnel_sock+0x6/0x4f
[   46.891355] *pdpt = 0000000000000000 *pde = f000ff53f000ff53
[   46.893918] Oops: 0002 [#1] PREEMPT
[   46.896014] CPU: 0 PID: 1 Comm: swapper Not tainted 4.7.0-rc4-00001-g8700e3e #1
[   46.900280] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
[   46.904905] task: cf06c040 ti: cf05e000 task.ti: cf05e000
[   46.907854] EIP: 0060:[<c220705a>] EFLAGS: 00210246 CPU: 0
[   46.911137] EIP is at setup_udp_tunnel_sock+0x6/0x4f
[   46.914070] EAX: 00000044 EBX: 00000001 ECX: cf05fef0 EDX: ca8142e0
[   46.917236] ESI: c2c4505b EDI: cf05fef0 EBP: cf05fed0 ESP: cf05fed0
[   46.919836]  DS: 007b ES: 007b FS: 0000 GS: 00e0 SS: 0068
[   46.922046] CR0: 80050033 CR2: 000001fc CR3: 02cec000 CR4: 000006b0
[   46.924550] Stack:
[   46.926014]  cf05ff10 c1fd4657 ca8142e0 0000000a 00000000 00000000 0000b712 00000008
[   46.931274]  00000000 6bb5bd01 c1fd48de 00000000 00000000 cf05ff1c 00000000 00000000
[   46.936122]  cf05ff1c c1fd4bdf 00000000 cf05ff28 c2c4507b ffffffff cf05ff88 c2bf1c74
[   46.942350] Call Trace:
[   46.944403]  [<c1fd4657>] rxe_setup_udp_tunnel+0x8f/0x99
[   46.947689]  [<c1fd48de>] ? net_to_rxe+0x4e/0x4e
[   46.950567]  [<c1fd4bdf>] rxe_net_init+0xe/0xa4
[   46.953147]  [<c2c4507b>] rxe_module_init+0x20/0x4c
[   46.955448]  [<c2bf1c74>] do_one_initcall+0x89/0x113
[   46.957797]  [<c2bf15eb>] ? set_debug_rodata+0xf/0xf
[   46.959966]  [<c2bf1dbc>] ? kernel_init_freeable+0xbe/0x15b
[   46.962262]  [<c2bf1ddc>] kernel_init_freeable+0xde/0x15b
[   46.964418]  [<c232eb54>] kernel_init+0x8/0xd0
[   46.966618]  [<c2333122>] ret_from_kernel_thread+0xe/0x24
[   46.969592]  [<c232eb4c>] ? rest_init+0x6f/0x6f

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Set source mac address in FTE

Set the source mac address in the FTE when L2 specification
is provided.

Fixes: 038d2ef87572 ('IB/mlx5: Add flow steering support')
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Enable MAD_IFC commands for IB ports only

MAD_IFC command is supported only for physical functions (PF)
and when physical port is IB. The proposed fix enforces it.

Fixes: d603c809ef91 ("IB/mlx5: Fix decision on using MAD_IFC")
Reported-by: David Chang <dchang@suse.com>
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Diagnostic HW counters are not supported in slave mode

Modify the mlx4_ib_diag_counters() to avoid the following error in the
hypervisor when the slave tries to query the hardware counters in SR-IOV
mode.

mlx4_core 0000:81:00.0: Unknown command:0x30 accepted from slave:1

Fixes: 3f85f2aaabf7 ("IB/mlx4: Add diagnostic hardware counters")
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Use correct subnet-prefix in QP1 mads under SR-IOV

When sending QP1 MAD packets which use a GRH, the source GID
(which consists of the 64-bit subnet prefix, and the 64 bit port GUID)
must be included in the packet GRH.

For SR-IOV, a GID cache is used, since the source GID needs to be the
slave's source GID, and not the Hypervisor's GID. This cache also
included a subnet_prefix. Unfortunately, the subnet_prefix field in
the cache was never initialized (to the default subnet prefix 0xfe80::0).
As a result, this field remained all zeroes.  Therefore, when SR-IOV
was active, all QP1 packets which included a GRH had a source GID
subnet prefix of all-zeroes.

However, the subnet-prefix should initially be 0xfe80::0 (the default
subnet prefix). In addition, if OpenSM modifies a port's subnet prefix,
the new subnet prefix must be used in the GRH when sending QP1 packets.
To fix this we now initialize the subnet prefix in the SR-IOV GID cache
to the default subnet prefix. We update the cached value if/when OpenSM
modifies the port's subnet prefix. We take this cached value when sending
QP1 packets when SR-IOV is active.

Note that the value is stored as an atomic64. This eliminates any need
for locking when the subnet prefix is being updated.

Note also that we depend on the FW generating the "port management change"
event for tracking subnet-prefix changes performed by OpenSM. If running
early FW (before 2.9.4630), subnet prefix changes will not be tracked (but
the default subnet prefix still will be stored in the cache; therefore
users who do not modify the subnet prefix will not have a problem).
IF there is a need for such tracking also for early FW, we will add that
capability in a subsequent patch.

Fixes: 1ffeb2eb8be9 ("IB/mlx4: SR-IOV IB context objects and proxy/tunnel SQP support")
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Fix code indentation in QP1 MAD flow

The indentation in the QP1 GRH flow in procedure build_mlx_header is
really confusing. Fix it, in preparation for a commit which touches
this code.

Fixes: 1ffeb2eb8be9 ("IB/mlx4: SR-IOV IB context objects and proxy/tunnel SQP support")
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Fix incorrect MC join state bit-masking on SR-IOV

Because of an incorrect bit-masking done on the join state bits, when
handling a join request we failed to detect a difference between the
group join state and the request join state when joining as send only
full member (0x8). This caused the MC join request not to be sent.
This issue is relevant only when SRIOV is enabled and SM supports
send only full member.

This fix separates scope bits and join states bits a nibble each.

Fixes: b9c5d6a64358 ('IB/mlx4: Add multicast group (MCG) paravirtualization for SR-IOV')
Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/ipoib: Don't allow MC joins during light MC flush

This fix solves a race between light flush and on the fly joins.
Light flush doesn't set the device to down and unset IPOIB_OPER_UP
flag, this means that if while flushing we have a MC join in progress
and the QP was attached to BC MGID we can have a mismatches when
re-attaching a QP to the BC MGID.

The light flush would set the broadcast group to NULL causing an on
the fly join to rejoin and reattach to the BC MCG as well as adding
the BC MGID to the multicast list. The flush process would later on
remove the BC MGID and detach it from the QP. On the next flush
the BC MGID is present in the multicast list but not found when trying
to detach it because of the previous double attach and single detach.

[18332.714265] ------------[ cut here ]------------
[18332.717775] WARNING: CPU: 6 PID: 3767 at drivers/infiniband/core/verbs.c:280 ib_dealloc_pd+0xff/0x120 [ib_core]
...
[18332.775198] Hardware name: Red Hat KVM, BIOS Bochs 01/01/2011
[18332.779411]  0000000000000000 ffff8800b50dfbb0 ffffffff813fed47 0000000000000000
[18332.784960]  0000000000000000 ffff8800b50dfbf0 ffffffff8109add1 0000011832f58300
[18332.790547]  ffff880226a596c0 ffff880032482000 ffff880032482830 ffff880226a59280
[18332.796199] Call Trace:
[18332.798015]  [<ffffffff813fed47>] dump_stack+0x63/0x8c
[18332.801831]  [<ffffffff8109add1>] __warn+0xd1/0xf0
[18332.805403]  [<ffffffff8109aebd>] warn_slowpath_null+0x1d/0x20
[18332.809706]  [<ffffffffa025d90f>] ib_dealloc_pd+0xff/0x120 [ib_core]
[18332.814384]  [<ffffffffa04f3d7c>] ipoib_transport_dev_cleanup+0xfc/0x1d0 [ib_ipoib]
[18332.820031]  [<ffffffffa04ed648>] ipoib_ib_dev_cleanup+0x98/0x110 [ib_ipoib]
[18332.825220]  [<ffffffffa04e62c8>] ipoib_dev_cleanup+0x2d8/0x550 [ib_ipoib]
[18332.830290]  [<ffffffffa04e656f>] ipoib_uninit+0x2f/0x40 [ib_ipoib]
[18332.834911]  [<ffffffff81772a8a>] rollback_registered_many+0x1aa/0x2c0
[18332.839741]  [<ffffffff81772bd1>] rollback_registered+0x31/0x40
[18332.844091]  [<ffffffff81773b18>] unregister_netdevice_queue+0x48/0x80
[18332.848880]  [<ffffffffa04f489b>] ipoib_vlan_delete+0x1fb/0x290 [ib_ipoib]
[18332.853848]  [<ffffffffa04df1cd>] delete_child+0x7d/0xf0 [ib_ipoib]
[18332.858474]  [<ffffffff81520c08>] dev_attr_store+0x18/0x30
[18332.862510]  [<ffffffff8127fe4a>] sysfs_kf_write+0x3a/0x50
[18332.866349]  [<ffffffff8127f4e0>] kernfs_fop_write+0x120/0x170
[18332.870471]  [<ffffffff81207198>] __vfs_write+0x28/0xe0
[18332.874152]  [<ffffffff810e09bf>] ? percpu_down_read+0x1f/0x50
[18332.878274]  [<ffffffff81208062>] vfs_write+0xa2/0x1a0
[18332.881896]  [<ffffffff812093a6>] SyS_write+0x46/0xa0
[18332.885632]  [<ffffffff810039b7>] do_syscall_64+0x57/0xb0
[18332.889709]  [<ffffffff81883321>] entry_SYSCALL64_slow_path+0x25/0x25
[18332.894727] ---[ end trace 09ebbe31f831ef17 ]---

Fixes: ee1e2c82c245 ("IPoIB: Refresh paths instead of flushing them on SM change events")
Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: fix GFP_KERNEL in spinlock context

There is skb_clone(skb, GFP_KERNEL) in spinlock context
in rxe_rcv_mcast_pkt().

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Acked-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Rework debugfs to use SRCU

The debugfs RCU trips many debug kernel warnings because of potential
sleeps with an RCU read lock held. This includes both user copy calls
and slab allocations throughout the file.

This patch switches the RCU to use SRCU for file remove/access
race protection.

In one case, the SRCU is implicit in the use of the raw debugfs file
object and just works.

In the seq_file case, a wrapper around seq_read() and seq_lseek() is
used to enforce the SRCU using the debugfs supplied functions
debugfs_use_file_start() and debugfs_use_file_stop().

The sychronize_rcu() is deleted since the SRCU prevents the remove
access race.

The RCU locking is kept for qp_stats since the QP hash list is
protected using the non-sleepable RCU.

Reviewed-by: Sebastian Sanchez <sebastian.sanchez@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: Make n_krcvqs be an unsigned long integer

The global variable n_krcvqs stores the sum of the number of kernel
receive queues of VLs 0-7 which the user can pass to the driver through
the module parameter array krcvqs which is of type unsigned integer. If
the user passes large value(s) into krcvqs parameter array, it can cause
an arithmetic overflow while calculating n_krcvqs which is also of type
unsigned int. The overflow results in an incorrect value of n_krcvqs
which can lead to kernel crash while loading the driver.

Fix by changing the data type of n_krcvqs to unsigned long. This patch
also changes the data type of other variables that get their values from
n_krcvqs.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add QSFP sanity pre-check

Sometimes a QSFP device does not respond in the expected time
after a power-on.  Add a read pre-check/retry when starting
the link on driver load.

Reviewed-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix AHG KDETH Intr shift

In the set_txreq_header_ahg(), The KDETH Intr bit is obtained from the
header in the user sdma request using a KDETH_GET shift and mask macro.
This value is then futher right shifted by 16 causing us to lose the
value i.e it is shifted to zero, leading to the following
smatch warning:
drivers/infiniband/hw/hfi1/user_sdma.c:1482 set_txreq_header_ahg()
warn: mask and shift to zero

The Intr bit should be left shifted into its correct position in the
KDETH header before the AHG update.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix SGE length for misaligned PIO copy

When trying to align the source pointer and there's a byte carry
in an SGE copy, bytes are borrowed from the next quad-word X to
complete the required quad-word copy. Then, the SGE length is
reduced by the number of borrowed bytes. After this, if the
remaining number of bytes from quad-word X (extra bytes) is
greater than the new SGE length, the number of extra bytes needs
to be updated to the new SGE length. Otherwise, when the
SGE length gets updated again after the extra bytes are read to
create the new byte carry, it goes negative, which then becomes
a very large number as the SGE length is an unsigned integer.
This causes SGE buffer to be over-read.

Reviewed-by: Dean Luick <dean.luick@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/mlx5: Don't return errors from poll_cq

Remove returning errors from mlx5 poll_cq function. Polling CQ
operation in kernel never fails by Mellanox HCA architecture and
respective driver design.

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

IB/mlx5: Use TIR number based on selector

Use TIR number based on selector, it should be done to differentiate
between RSS QP to RAW one.

Reported-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Tested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Simplify code by removing return variable

Return variable was set in a line before the
actual return was called in begin_wqe function.

This patch removes such variable and simplifies the code.

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

IB/mlx5: Return EINVAL when caller specifies too many SGEs

The returned value should be EINVAL, because it is caused by wrong
caller and not by internal overflow event.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Don't return errors from poll_cq

Remove returning errors from mlx4 poll_cq function. Polling CQ
operation in kernel never fails by Mellanox HCA architecture and
respective driver design.

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

Revert "IB/mlx4: Return EAGAIN for any error in mlx4_ib_poll_one"

By Mellanox HW design and SW implementation, poll_cq never
fails and returns errors, so all these printks are to catch ULP bugs.

In case of such bug, the reverted patch will cause reentry of the
function, resulting in a printk storm.

This reverts commit 5412352fcd8f ("IB/mlx4: Return EAGAIN for any error in mlx4_ib_poll_one")

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

IB/ipoib: Fix memory corruption in ipoib cm mode connect flow

When a new CM connection is being requested, ipoib driver copies data
from the path pointer in the CM/tx object, the path object might be
invalid at the point and memory corruption will happened later when now
the CM driver will try using that data.

The next scenario demonstrates it:
neigh_add_path --> ipoib_cm_create_tx -->
queue_work (pointer to path is in the cm/tx struct)
#while the work is still in the queue,
#the port goes down and causes the ipoib_flush_paths:
ipoib_flush_paths --> path_free --> kfree(path)
#at this point the work scheduled starts.
ipoib_cm_tx_start --> copy from the (invalid)path pointer:
(memcpy(&pathrec, &p->path->pathrec, sizeof pathrec);)
 -> memory corruption.

To fix that the driver now starts the CM/tx connection only if that
specific path exists in the general paths database.
This check is protected with the relevant locks, and uses the gid from
the neigh member in the CM/tx object which is valid according to the ref
count that was taken by the CM/tx.

Fixes: 839fcaba35 ('IPoIB: Connected mode experimental support')
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Fix use after free in send_leave function

The function send_leave sets the member: group->query_id
(group->query_id = ret) after calling the sa_query, but leave_handler
can be executed before the setting and it might delete the group object,
and will get a memory corruption.

Additionally, this patch gets rid of group->query_id variable which is
not used.

Fixes: faec2f7b96b5 ('IB/sa: Track multicast join/leave requests')
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/cxgb4: Make _free_qp static to silence build warning

We get 1 warning when build kernel with W=1:
drivers/infiniband/hw/cxgb4/qp.c:686:6: warning: no previous prototype for '_free_qp' [-Wmissing-prototypes]

In fact, this function is only used in the file in which it is declared
and don't need a declaration, but can be made static.
so this patch marks it 'static'.

Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/isert: Properly release resources on DEVICE_REMOVAL

When the low level driver exercises the hot unplug they would call
rdma_cm cma_remove_one which would fire DEVICE_REMOVAL event to all cma
consumers. Now, if consumer doesn't make sure they destroy all IB
objects created on that IB device instance prior to finalizing all
processing of DEVICE_REMOVAL callback, rdma_cm will let the lld to
de-register with IB core and destroy the IB device instance. And if the
consumer calls (say) ib_dereg_mr(), it will crash since that dev object
is NULL.

In the current implementation, iser-target just initiates the cleanup
and returns from DEVICE_REMOVAL callback. This deferred work creates a
race between iser-target cleaning IB objects(say MR) and lld destroying
IB device instance.

This patch includes the following fixes
  -> make sure that consumer frees all IB objects associated with device
     instance
  -> return non-zero from the callback to destroy the rdma_cm id

Signed-off-by: Raju Rangoju <rajur@chelsio.com>
Acked-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix the size parameter to find_first_bit

The 2nd parameter of 'find_first_bit' is the number of bits to search.
In this case, we are passing 'sizeof(u64)' which is 8.

It is likely that the number of bits of 'port_mask' was expected here.
Use sizeof() * 8 to get the correct number.

It has been spotted by the following coccinelle script:
@@
expression ret, x;

@@
*  ret = \(find_first_bit \| find_first_zero_bit\) (x, sizeof(...));

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Fix the size parameter to find_first_bit

The 2nd parameter of 'find_first_bit' is the number of bits to search.
In this case, we are passing 'sizeof(tmp)' which is likely to be 4 or 8
because 'tmp' is an 'unsigned long'.

It is likely that the number of bits of 'tmp' was expected here. So use
BITS_PER_LONG instead.

It has been spotted by the following coccinelle script:
@@
expression ret, x;

@@
*  ret = \(find_first_bit \| find_first_zero_bit\) (x, sizeof(...));

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Acked-by: Majd Dibbiny <majd@mellanox.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Clean up type used and casting

In all other places in this file where 'find_first_bit' is called,
port_num is defined as a 'u8' and no casting is done.

Do the same here in order to be more consistent.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Receive notification events correctly

Device notifications are not received after the first interface is
closed; since there is an unregister for notifications on every
interface close. Correct this by unregistering for device
notifications only when the last interface is closed. Also, make
all operations on the i40iw_notifiers_registered atomic as it
can be read/modified concurrently.

Fixes: 8e06af711bf2 ("i40iw: add main, hdr, status")
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: Update hw_iwarp_state

Update iwqp->hw_iwarp_state to reflect the new state of the CQP
modify QP operation. This avoids reissuing a CQP operation to
modify a QP to a state that it is already in.

Fixes: 4e9042e647ff ("i40iw: add hw and utils files")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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 'misc-fixes' into k.o/for-4.8-rc

i40iw: Send last streaming mode message for loopback connections

Send a zero length last streaming mode message for loopback
connections to synchronize between accepting QP and connecting QP.
This avoids data transfer to start on the accepting QP before
the connecting QP is in RTS. Also remove function i40iw_loopback_nop()
as it is no longer used.

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: Doug Ledford <dledford@redhat.com>

IB/srpt: Update sport->port_guid with each port refresh

If port_guid is set with the default subnet_prefix, then we get a change
event and run a port refresh, we don't update the port_guid.  As a
result, attempts to create a target device that uses the new
subnet_prefix in the wwn will fail to find a match and be rejected by
the ib_srpt driver.  This makes it impossible to configure a port if it
was initialized with a default subnet_prefix and later changed to any
non-default subnet-prefix.  Updating the port refresh task to always
update the wwn based upon the current subnext_prefix solves this
problem.

Cc: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: nab@linux-iscsi.org
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/ocrdma: Fix the max_sge reported from FW

Current driver is reporting wrong values for max_sge and
max_sge_rd in query_device. This breaks the nfs rdma and iser
in some device profiles. Fixing the driver to report
correct values from FW.

Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Avoid writing to freed memory

iwpbl->iwmr points to the structure that contains iwpbl,
which is iwmr. Setting this to NULL would result in
writing to freed memory. So just free iwmr, and return.

Fixes: d37498417947 ("i40iw: add files for iwarp interface")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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: Fix double free of allocated_buffer

Memory allocated for iwqp; iwqp->allocated_buffer is freed twice in
the create_qp error path. Correct this by having it freed only once in
i40iw_free_qp_resources().

Fixes: d37498417947 ("i40iw: add files for iwarp interface")
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>

IB/mlx5: Remove superfluous include of io-mapping.h

This file does not use any structs or functions defined by io-mapping.h
(nor does it directly use iomap, ioremap, iounamp or friends). Remove it
to simplify verification of changes to io-mapping.h

The include existed since its inception in

commit e126ba97dba9edeb6fafa3665b5f8497fc9cdf8c
Author: Eli Cohen <eli@mellanox.com>
Date:   Sun Jul 7 17:25:49 2013 +0300

    mlx5: Add driver for Mellanox Connect-IB adapters

which looks like a copy across from the Mellanox ethernet driver.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Eli Cohen <eli@mellanox.com>
Cc: Jack Morgenstein <jackm@dev.mellanox.co.il>
Cc: Or Gerlitz <ogerlitz@mellanox.com>
Cc: Matan Barak <matanb@mellanox.com>
Cc: Leon Romanovsky <leonro@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
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Laurence Oberman <loberman@redhat.com>
Tested-by: Laurence Oberman <loberman@redhat.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Do not set self-referencing pointer to NULL after kfree

In i40iw_free_virt_mem(), do not set mem->va to NULL
after freeing it as mem->va is a self-referencing pointer
to mem.

Fixes: 4e9042e647ff ("i40iw: add hw and utils files")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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: Add missing NULL check for MPA private data

Add NULL check for pdata and pdata->addr before the memcpy in
i40iw_form_cm_frame(). This fixes a NULL pointer de-reference
which occurs when the MPA private data pointer is NULL. Also
only copy pdata->size bytes in the memcpy to prevent reading
past the length of the private data buffer provided by upper layer.

Fixes: f27b4746f378 ("i40iw: add connection management code")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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>

iw_cxgb4: Fix cxgb4 arm CQ logic w/IB_CQ_REPORT_MISSED_EVENTS

Current cxgb4 arm CQ logic ignores IB_CQ_REPORT_MISSED_EVENTS for
request completion notification on a CQ. Due to this ib_poll_handler()
assumes all events polled and avoids further iopoll scheduling.

This patch adds logic to cxgb4 ib_req_notify_cq() handler to check if
CQ is not empty and return accordingly. Based on the return value of
ib_req_notify_cq() handler, ib_poll_handler() will schedule a run of
iopoll handler.

Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Add missing check for interface already open

In i40iw_open(), check if interface is already open
and return success if it is.

Fixes: 8e06af711bf2 ("i40iw: add main, hdr, status")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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: Protect req_resource_num update

In i40iw_alloc_resource(), ensure that the update to
req_resource_num is protected by the lock.

Fixes: 8e06af711bf2 ("i40iw: add main, hdr, status")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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: Change mem_resources pointer to a u8

iwdev->mem_resources is incorrectly defined as an unsigned
long instead of u8. As a result, the offset into the dynamic
allocated structures in i40iw_initialize_hw_resources() is
incorrectly calculated and would lead to writing of memory
regions outside of the allocated buffer.

Fixes: 8e06af711bf2 ("i40iw: add main, hdr, status")
Reported-by: Stefan Assmann <sassmann@redhat.com>
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>

IB/core: Use memdup_user() rather than duplicating its implementation

* Reuse existing functionality from memdup_user() instead of keeping
  duplicate source code.

  This issue was detected by using the Coccinelle software.

* The local variable "ret" will be set to an appropriate value a bit later.
  Thus omit the explicit initialisation at the beginning.

Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/qib: Use memdup_user() rather than duplicating its implementation

Reuse existing functionality from memdup_user() instead of keeping
duplicate source code.

This issue was detected by using the Coccinelle software.

Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

iw_cxgb4: use the MPA initiator's IRD if < our ORD

The i40iw initiator sends an MPA-request with ird=16 and ord=16. The cxgb4
responder sends an MPA-reply with ord = 32 causing i40iw to terminate
due to insufficient resources.

The logic to reduce the ORD to <= peer's IRD was wrong.

Reported-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

iw_cxgb4: limit IRD/ORD advertised to ULP by device max.

The i40iw initiator sends an MPA-request with ird = 63, ord = 63. The
cxgb4 responder sends a RST.  Since the inbound ord=63 and it exceeds
the max_ird/c4iw_max_read_depth (=32 default), chelsio decides to abort.

Instead, cxgb4 should adjust the ord/ird down before presenting it to
the ULP.

Reported-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fix mm_struct use after free

Testing with CONFIG_SLUB_DEBUG_ON=y resulted in the kernel panic below.

This is the result of the mm_struct sometimes being free'd prior to
hfi1_file_close being called.

This was due to the combination of 2 reasons:

1) hfi1_file_close is deferred in process exit and it therefore may not
   be called synchronously with process exit.
2) exit_mm is called prior to exit_files in do_exit.  Normally this is ok
   however, our kernel bypass code requires us to have access to the
   mm_struct for house keeping both at "normal" close time as well as at
   process exit.

Therefore, the fix is to simply keep a reference to the mm_struct until
we are done with it.

[ 3006.340150] general protection fault: 0000 [#1] SMP
[ 3006.346469] Modules linked in: hfi1 rdmavt rpcrdma ib_isert iscsi_target_mod
ib_iser libiscsi scsi_transport_iscsi ib_srpt target_core_mod
 ib_srp scsi_transport_srp 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 snd_hda_code
 c_realtek iTCO_wdt snd_hda_codec_generic iTCO_vendor_support sb_edac edac_core
 x86_pkg_temp_thermal intel_powerclamp coretemp kvm irqbypass c
 rct10dif_pclmul crc32_pclmul ghash_clmulni_intel aesni_intel lrw snd_hda_intel
 gf128mul snd_hda_codec glue_helper snd_hda_core ablk_helper sn
 d_hwdep cryptd snd_seq snd_seq_device snd_pcm snd_timer snd soundcore pcspkr
 shpchp mei_me sg lpc_ich mei i2c_i801 mfd_core ioatdma ipmi_devi
 ntf wmi ipmi_si ipmi_msghandler acpi_cpufreq nfsd auth_rpcgss nfs_acl lockd
 grace sunrpc ip_tables ext4 jbd2 mbcache mlx4_en ib_core sr_mod s
 d_mod cdrom crc32c_intel mgag200 drm_kms_helper syscopyarea sysfillrect igb
 sysimgblt fb_sys_fops ptp mlx4_core ttm isci pps_core ahci drm li
 bsas libahci dca firewire_ohci i2c_algo_bit scsi_transport_sas firewire_core
 crc_itu_t i2c_core libata [last unloaded: mlx4_ib]
 [ 3006.461759] CPU: 16 PID: 11624 Comm: mpi_stress Not tainted 4.7.0-rc5+ #1
 [ 3006.469915] Hardware name: Intel Corporation W2600CR ........../W2600CR, BIOS SE5C600.86B.01.08.0003.022620131521 02/26/2013
 [ 3006.483027] task: ffff8804102f0040 ti: ffff8804102f8000 task.ti: ffff8804102f8000
 [ 3006.491971] RIP: 0010:[<ffffffff810f0383>]  [<ffffffff810f0383>] __lock_acquire+0xb3/0x19e0
 [ 3006.501905] RSP: 0018:ffff8804102fb908  EFLAGS: 00010002
 [ 3006.508447] RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000001 RCX: 0000000000000000
 [ 3006.517012] RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff880410b56a40
 [ 3006.525569] RBP: ffff8804102fb9b0 R08: 0000000000000001 R09: 0000000000000000
 [ 3006.534119] R10: ffff8804102f0040 R11: 0000000000000000 R12: 0000000000000000
 [ 3006.542664] R13: ffff880410b56a40 R14: 0000000000000000 R15: 0000000000000000
 [ 3006.551203] FS:  00007ff478c08700(0000) GS:ffff88042e200000(0000) knlGS:0000000000000000
 [ 3006.560814] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 [ 3006.567806] CR2: 00007f667f5109e0 CR3: 0000000001c06000 CR4: 00000000000406e0
 [ 3006.576352] Stack:
 [ 3006.579157]  ffffffff8124b819 ffffffffffffffff 0000000000000000 ffff8804102fb940
 [ 3006.588072]  0000000000000002 0000000000000000 ffff8804102f0040 0000000000000007
 [ 3006.596971]  0000000000000006 ffff8803cad6f000 0000000000000000 ffff8804102f0040
 [ 3006.605878] Call Trace:
 [ 3006.609220]  [<ffffffff8124b819>] ? uncharge_batch+0x109/0x250
 [ 3006.616382]  [<ffffffff810f2313>] lock_acquire+0xd3/0x220
 [ 3006.623056]  [<ffffffffa0a30bfc>] ? hfi1_release_user_pages+0x7c/0xa0 [hfi1]
 [ 3006.631593]  [<ffffffff81775579>] down_write+0x49/0x80
 [ 3006.638022]  [<ffffffffa0a30bfc>] ? hfi1_release_user_pages+0x7c/0xa0 [hfi1]
 [ 3006.646569]  [<ffffffffa0a30bfc>] hfi1_release_user_pages+0x7c/0xa0 [hfi1]
 [ 3006.654898]  [<ffffffffa0a2efb6>] cacheless_tid_rb_remove+0x106/0x330 [hfi1]
 [ 3006.663417]  [<ffffffff810efd36>] ? mark_held_locks+0x66/0x90
 [ 3006.670498]  [<ffffffff817771f6>] ? _raw_spin_unlock_irqrestore+0x36/0x60
 [ 3006.678741]  [<ffffffffa0a2f1ee>] tid_rb_remove+0xe/0x10 [hfi1]
 [ 3006.686010]  [<ffffffffa0a0c5d5>] hfi1_mmu_rb_unregister+0xc5/0x100 [hfi1]
 [ 3006.694387]  [<ffffffffa0a2fcb9>] hfi1_user_exp_rcv_free+0x39/0x120 [hfi1]
 [ 3006.702732]  [<ffffffffa09fc6ea>] hfi1_file_close+0x17a/0x330 [hfi1]
 [ 3006.710489]  [<ffffffff81263e9a>] __fput+0xfa/0x230
 [ 3006.716595]  [<ffffffff8126400e>] ____fput+0xe/0x10
 [ 3006.722696]  [<ffffffff810b95c6>] task_work_run+0x86/0xc0
 [ 3006.729379]  [<ffffffff81099933>] do_exit+0x323/0xc40
 [ 3006.735672]  [<ffffffff8109a2dc>] do_group_exit+0x4c/0xc0
 [ 3006.742371]  [<ffffffff810a7f55>] get_signal+0x345/0x940
 [ 3006.748958]  [<ffffffff810340c7>] do_signal+0x37/0x700
 [ 3006.755328]  [<ffffffff8127872a>] ? poll_select_set_timeout+0x5a/0x90
 [ 3006.763146]  [<ffffffff811609cb>] ? __audit_syscall_exit+0x1db/0x260
 [ 3006.770853]  [<ffffffff8110f3e3>] ? rcu_read_lock_sched_held+0x93/0xa0
 [ 3006.778765]  [<ffffffff812347a4>] ? kfree+0x1e4/0x2a0
 [ 3006.784986]  [<ffffffff8108e75a>] ? exit_to_usermode_loop+0x33/0xac
 [ 3006.792551]  [<ffffffff8108e785>] exit_to_usermode_loop+0x5e/0xac
 [ 3006.799907]  [<ffffffff81003dca>] do_syscall_64+0x12a/0x190
 [ 3006.806664]  [<ffffffff81777a7f>] entry_SYSCALL64_slow_path+0x25/0x25
 [ 3006.814396] Code: 24 08 44 89 44 24 10 89 4c 24 18 e8 a8 d8 ff ff 48 85 c0
 8b 4c 24 18 44 8b 44 24 10 44 8b 4c 24 08 4c 8b 14 24 0f 84 30
 08 00 00 <f0> ff 80 98 01 00 00 8b 3d 48 ad be 01 45 8b a2 90 0b 00 00 85
 [ 3006.837158] RIP  [<ffffffff810f0383>] __lock_acquire+0xb3/0x19e0
 [ 3006.844401]  RSP <ffff8804102fb908>
 [ 3006.851170] ---[ end trace b7b9f21cf06c27df ]---
 [ 3006.927420] Kernel panic - not syncing: Fatal exception
 [ 3006.933954] Kernel Offset: disabled
 [ 3006.940961] ---[ end Kernel panic - not syncing: Fatal exception
 [ 3006.948249] ------------[ cut here ]------------

Fixes: 3faa3d9a308e ("IB/hfi1: Make use of mm consistent")
Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rdmvat: Fix double vfree() in rvt_create_qp() error path

The unwind logic for creating a user QP has a double vfree
of the non-shared receive queue when handling a "too many qps"
failure.

The code unwinds the mmmap info by decrementing a reference
count which will call rvt_release_mmap_info() which in turn
does the vfree() of the r_rq.wq.  The unwind code then does
the same free.

Fix by guarding the vfree() with the same test that is done
in close and only do the vfree() if qp->ip is NULL.

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: Improve J_KEY generation

Previously, J_KEY generation was based on the lower 16 bits
of the user's UID. While this works, it was not good enough
as a non-root user could collide with a root user given a
sufficiently large UID.

This patch attempt to improve the J_KEY generation by using
the following algorithm:

The 16 bit J_KEY space is partitioned into 3 separate spaces
reserved for different user classes:
   * all users with administtor privileges (including 'root')
     will use J_KEYs in the range of 0 to 31,
   * all kernel protocols, which use KDETH packets will use
     J_KEYs in the range of 32 to 63, and
   * all other users will use J_KEYs in the range of 64 to
     65535.

The above separation is aimed at preventing different user levels
from sending packets to each other and, additionally, separate
kernel protocols from all other types of users. The later is meant
to prevent the potential corruption of kernel memory by any other
type of user.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Return invalid field for non-QSFP CableInfo queries

The driver does not check if the CableInfo query is supported for the
port type. Return early if CableInfo is not supported for the port type,
making compliance with the specification explicit and preventing lower
level code from potentially doing the wrong thing if the query is not
supported for the hardware implementation.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

MAINTAINERS: Fix Soft RoCE location

The Soft RoCE (rxe) is located in drivers/inifiniband/sw
and not in drivers/infiniband/hw/.

This patch fixes it.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/usnic: Fix error return code

If 'pci_register_driver' fails, we return 'err' which is known to be 0.
Return the error instead.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Add missing error code assignment before test

It is likely that checking the result of 'setup_ctxt' is expected here.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Using kfree_rcu() to simplify the code

The callback function of call_rcu() just calls a kfree(), so we
can use kfree_rcu() instead of call_rcu() + callback function.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Tested-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Acked-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Tested-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Acked-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Validate header in set_armed_active

Validate the etype to insure that the header is correct.

Reviewed-by: Don Hiatt <don.hiatt@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Pass packet ptr to set_armed_active

The "packet" parameter was being passed on the stack,
change it to a pointer.

Reviewed-by: Don Hiatt <don.hiatt@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Fetch monitor values on-demand for CableInfo query

The monitor values from bytes 22 through 81 of the QSFP memory space
(SFF 8636) are dynamic and serving them out of the QSFP memory cache
maintained by the driver provides stale data to the CableInfo SMA query.
This patch refreshes the dynamic values from the QSFP memory on request
and overwrites the stale data from the cache for the overlap between the
requested range and the monitor range.

Reviewed-by: Jubin John <jubin.john@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Easwar Hariharan <easwar.hariharan@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1,IB/qib: Fix qp_stats sleep with rcu read lock held

The qp init function does a kzalloc() while holding the RCU
lock that encounters the following warning with a debug kernel
when a cat of the qp_stats is done:

[  231.723948] rcu_scheduler_active = 1, debug_locks = 0
[  231.731939] 3 locks held by cat/11355:
[  231.736492]  #0:  (debugfs_srcu){......}, at: [<ffffffff813001a5>] debugfs_use_file_start+0x5/0x90
[  231.746955]  #1:  (&p->lock){+.+.+.}, at: [<ffffffff81289a6c>] seq_read+0x4c/0x3c0
[  231.755873]  #2:  (rcu_read_lock){......}, at: [<ffffffffa0a0c535>] _qp_stats_seq_start+0x5/0xd0 [hfi1]
[  231.766862]

The init functions do an implicit next which requires the rcu read lock
before the kzalloc().

Fix for both drivers is to change the scope of the init function to only
do the allocation and the initialization of the just allocated iter.

The implict next is moved back into the respective start functions to fix
the issue.

Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
CC: <stable@vger.kernel.org> # 4.6.x-
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Remove duplicated include from affinity.c

Remove duplicated include.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>