RDMA/core: Introduce IB_MR_TYPE_INTEGRITY and ib_alloc_mr_integrity API

This is a preparation for signature verbs API re-design. In the new
design a single MR with IB_MR_TYPE_INTEGRITY type will be used to perform
the needed mapping for data integrity operations.

Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Save the MR type in the ib_mr structure

This is a preparation for the signature verbs API change. This change is
needed since the MR type will define, in the upcoming patches, the need
for allocating internal resources in LLD for signature handover related
operations. It will also help to make sure that signature related
functions are called with an appropriate MR type and fail otherwise.
Also introduce new mr types IB_MR_TYPE_USER, IB_MR_TYPE_DMA and
IB_MR_TYPE_DM for correctness.

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Introduce new header file for signature operations

Ease the exhausted ib_verbs.h file and make the code more readable.

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/efa: Print address on AH creation failure

For debugging purposes, print destination address if failed to create AH.

Signed-off-by: Firas Jahjah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/efa: Be consistent with success flow return value

The EFA driver is written with success oriented flows in mind, meaning
that functions should mostly end with a return 0 statement.
Error flows return their error value on their own instead of assuming
that the function will return the error at the end.

This commit fixes a bunch of functions that were not aligned with this
behavior.

Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/efa: Use API to get contiguous memory blocks aligned to device supported page size

Use the ib_umem_find_best_pgsz() and rdma_for_each_block() API when
registering an MR instead of coding it in the driver.

ib_umem_find_best_pgsz() is used to find the best suitable page size
which replaces the existing efa_cont_pages() implementation.
rdma_for_each_block() is used to iterate the umem in aligned contiguous
memory blocks.

Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/{rdmavt, qib, hfi1}: Convert to new completion API

Convert all completions to use the new completion routine that
fixes a race between post send and completion where fields from
a SWQE can be read after SWQE has been freed.

This patch also addresses issues reported in
https://marc.info/?l=linux-kernel&m=155656897409107&w=2.

The reserved operation path has no need for any barrier.

The barrier for the other path is addressed by the
smp_load_acquire() barrier.

Cc: Andrea Parri <andrea.parri@amarulasolutions.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@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 new completion inline

There is opencoded send completion logic all over all
the drivers.

We need to convert to this routine to enforce ordering
issues for completions.  This routine fixes an ordering
issue where the read of the SWQE fields necessary for creating
the completion can race with a post send if the post send catches
a send queue at the edge of being full.  Is is possible in that situation
to read SWQE fields that are being written.

This new routine insures that SWQE fields are read prior to advancing
the index that post send uses to determine queue fullness.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@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>

RDMA/odp: Do not leak dma maps when working with huge pages

The ib_dma_unmap_page() must match the length of the ib_dma_map_page(),
which is based on odp_shift. Otherwise iommu resources under this API
will not be properly freed.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/uverbs: Use offsetofend instead of opencoding

Discovered this was available already.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMa/hns: Don't stuck in endless timeout loop

The "end" variable is declared as unsigned and can't be negative, it
leads to the situation where timeout limit is not honored, so let's
convert logic to ensure that loop is bounded.

drivers/infiniband/hw/hns/hns_roce_hw_v1.c: In function _hns_roce_v1_clear_hem_:
drivers/infiniband/hw/hns/hns_roce_hw_v1.c:2471:12: warning: comparison of unsigned expression < 0 is always false [-Wtype-limits]
 2471 |    if (end < 0) {
      |            ^

Fixes: 669cefb654cb ("RDMA/hns: Remove jiffies operation in disable interrupt context")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA: Check umem pointer validity prior to release

Update ib_umem_release() to behave similarly to kfree() and allow
submitting NULL pointer as safe input to this function.

Fixes: a52c8e2469c3 ("RDMA: Clean destroy CQ in drivers do not return errors")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: reset function when removing module

During removing the driver, we needs to notify the roce engine to
stop working immediately,and symmetrically recycle the hardware
resources requested during initialization.

The hardware provides a command called function clear that can package
these operations,so that the driver can only focus on releasing
resources that applied from the operating system.
This patch implements the call of this command.

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

RDMA: Convert destroy_wq to be void

All callers of destroy WQ are always success and there is no need
to check their return value, so convert destroy_wq to be void.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Fix bug when wqe num is larger than 16K

hip08 can support up to 32768 wqes in one qp. currently if the wqe num
is larger than 16384, the driver will lead a calltrace as follows.

[21361.393725] Call trace:
[21361.398605]  hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2]
[21361.410627]  hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce]
[21361.420906]  _ib_modify_qp+0x70/0x118
[21361.428222]  ib_modify_qp+0x14/0x1c
[21361.435193]  rt_ktest_modify_qp+0xb8/0x650 [rdma_test]
[21361.445472]  exec_modify_qp_cmd+0x110/0x4d8 [rdma_test]
[21361.455924]  rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test]
[21361.467422]  rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test]
[21361.478570]  rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test]
[21361.489545]  rt_ktest_dev_write+0x328/0x4b0 [rdma_test]
[21361.499998]  __vfs_write+0x38/0x15c
[21361.506966]  vfs_write+0xa8/0x1a0
[21361.513586]  ksys_write+0x50/0xb0
[21361.520206]  sys_write+0xc/0x14
[21361.526479]  el0_svc_naked+0x30/0x34
[21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0)
[21361.545815] ---[ end trace e2a1feb2c3d7f13c ]---

When the wqe num is larger than 16384, hns_roce_table_find will return an
invalid mtt, this will lead an kernel paging requet error if the driver try
to access it. It's the mtt design defect which can't support up to the max
wqe num of hip08.

This patch fixs it by replacing mtt with mtr for wqe.

Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Add a group interfaces for optimizing buffers getting flow

Currently, the code for getting umem and kmem buffers exist many files,
this patch adds a group interfaces to simplify the buffers getting flow.

Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/hns: Add mtr support for mixed multihop addressing

Currently, the MTT(memory translate table) design required a buffer
space must has the same hopnum, but the hip08 hw can support mixed
hopnum config in a buffer space.

This patch adds the MTR(memory translate region) design for supporting
mixed multihop.

Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/netlink: Resort policy array

Sort the netlink policy array by netlink attribute name.  This will make
it easier in the future to find the entry you are looking for when you
need to make changes, or to make sure you don't add the same entry
twice.

Fix the whitespace while we are there.

Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/mlx5: Enable decap and packet reformat on FDB

If FDB flow tables support decap operation, enable it on creation,
This allows to perform decapsulation of tunnelled packets by steering
rules. If FDB flow tables support reformat operation, enable it on
creation as well.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/mlx5: Consider eswitch encap mode

When flow steering is created, then the encap support should
consider the eswitch encap mode. If the eswitch flow table (FDB)
supports encap then it shouldn't be supported on NIC RX flow tables.

Fixes: 4adda1122c490 ('RDMA/mlx5: Enable decap and packet reformat on flow tables')
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge remote-tracking branch 'mlx5-next/mlx5-next' into HEAD

Take mlx5-next so we can take a dependent two patch series next.

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

RDMA/odp: Fix missed unlock in non-blocking invalidate_start

If invalidate_start returns with EAGAIN then the umem_rwsem needs to be
unlocked as no invalidate_end will be called.

Cc: <stable@vger.kernel.org>
Fixes: ca748c39ea3f ("RDMA/umem: Get rid of per_mm->notifier_count")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Spelling s/statisfied/satisfied/

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA: Report available cdevs through RDMA_NLDEV_CMD_GET_CHARDEV

Update the struct ib_client for all modules exporting cdevs related to the
ibdevice to also implement RDMA_NLDEV_CMD_GET_CHARDEV. All cdevs are now
autoloadable and discoverable by userspace over netlink instead of relying
on sysfs.

uverbs also exposes the DRIVER_ID for drivers that are able to support
driver id binding in rdma-core.

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

RDMA: Add NLDEV_GET_CHARDEV to allow char dev discovery and autoload

Allow userspace to issue a netlink query against the ib_device for
something like "uverbs" and get back the char dev name, inode major/minor,
and interface ABI information for "uverbs0".

Since we are now in netlink this can also trigger a module autoload to
make the uverbs device come into existence.

Largely this will let us replace searching and reading inside sysfs to
setup devices, and provides an alternative (using driver_id) to device
name based provider binding for things like rxe.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA: Move rdma_node_type to uapi/

This enum is exposed over the sysfs file 'node_type' and over netlink via
RDMA_NLDEV_ATTR_DEV_NODE_TYPE, so declare it in the uapi headers.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

net/mlx5: Expose eswitch encap mode

Add API to get the current Eswitch encap mode.
It will be used in downstream patches to check if
flow table can be created with encap support or not.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>

net/mlx5: Declare more strictly devlink encap mode

Devlink has UAPI declaration for encap mode, so there is no
need to be loose on the data get/set by drivers.

Update call sites to use enum devlink_eswitch_encap_mode
instead of plain u8.

Suggested-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Petr Vorel <pvorel@suse.cz>

net/mlx5: Add EQ enable/disable API

Previously, EQ joined the chain notifier on creation.
This forced the caller to be ready to handle events before creating
the EQ through eq_create_generic interface.

To help the caller control when the created EQ will be attached to the
IRQ, add enable/disable API.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Use a single IRQ for all async EQs

The patch modifies the IRQ allocation so that all async EQs are
assigned to the same IRQ resulting in more available IRQs for
completion EQs.

The changes are using the support for IRQ sharing and EQ polling budget
that was introduced in previous patches so when the shared interrupt is
triggered, the kernel will serially call the handler of each of the
sharing EQs with a certain budget of EQEs to poll in order to prevent
starvation.

Signed-off-by: Ariel Levkovich <lariel@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Rename mlx5_irq_info to mlx5_irq

struct mlx5_irq_info is an active object and not just info.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Move all IRQ logic to pci_irq.c

Finalize IRQ separation and expose irq interface.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Generalize IRQ interface to work with irq_table

IRQ interface should operate within the irq_table context.
It should be independent of any EQ data structure.

The interface that will be exposed:
init/clenup, create/destroy, attach/detach

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Separate IRQ table creation from EQ table creation

IRQ allocation should be part of the IRQ table life-cycle.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Move IRQ affinity set to IRQ allocation phase

Affinity set/clear is part of the IRQ life-cycle.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Move IRQ rmap creation to IRQ allocation phase

Rmap creation/deletion is part of the IRQ life-cycle.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Separate IRQ data from EQ table data

IRQ table should only exist for mlx5_core_dev for PF and VF only.
EQ table of mediated devices should hold a pointer to the IRQ table
of the parent PCI device.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Separate IRQ request/free from EQ life cycle

Instead of requesting IRQ with eq creation, IRQs will be requested
before EQ table creation.
Instead of freeing the IRQs after EQ destroy, free IRQs after eq
table destroy.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Change interrupt handler to call chain notifier

Multiple EQs may share the same IRQ in subsequent patches.

Instead of calling the IRQ handler directly, the EQ will register
to an atomic chain notfier.

The Linux built-in shared IRQ is not used because it forces the caller
to disable the IRQ and clear affinity before free_irq() can be called.

This patch is the first step in the separation of IRQ and EQ logic.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Introduce EQ polling budget

Multiple EQs may share the same irq in subsequent patches.
To avoid starvation, a budget is set per EQ's interrupt handler.

Because of this change, it is no longer required to check that
MLX5_NUM_SPARE_EQE eqes were polled (to detect that arm is required).
It is guaranteed that MLX5_NUM_SPARE_EQE > budget, therefore the
handler will arm and exit the handler before all the entries in the
eq are polled.

In the scenario where the handler is out of budget and there are more
EQEs to poll, arming the EQ guarantees that the HW will send another
interrupt and the handler will be called again.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Support querying max VFs from device

For ECPF with eswitch manager privilege, query the host max VF count
by querying the device using query_functions command.

With this enhancement:
1. flow steering entries are created only for valid vports based on
   the max VF count of the PF.
2. Driver only queries cap of valid vport.

Eswitch requires the max VFs when doing initialization, so do sr-iov
init before eswitch init.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: E-Switch, Return raw output for query esw functions

Current function only returns host num of VFs, later patch requires
other params such as host maximum num of VFs.

Return the raw output so that caller can extract info as needed.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: E-Switch, Handle representors creation in handler context

Unified representors creation in esw_functions_changed context
handler. Emulate the esw_function_changed event for FW/HW that
does not support this event.

Signed-off-by: Vu Pham <vuhuong@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Increase wait time for fw initialization

Firmware FLR happens sequentially, in some cases, like when destroying
a VM that had many VFs, may require waiting much longer than 10 seconds.
Increase the timeout to 2 minutes, and print a wait countdown status
every 20 seconds.

Signed-off-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

rdma: Remove nes

This driver was first merged over 10 years ago and has not seen major
activity by the authors in the last 7 years. However, in that time it has
been patched 150 times to adapt it to changing kernel APIs.

Further, the hardware has several issues, like not supporting 64 bit DMA,
that make it rather uninteresting for use with modern systems and RDMA.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/ipoib: Remove check for ETH_SS_TEST

The default action for unlisted tests is "not-supported", so given that
ipoib doesn't support ETH_SS_TEST, there is no need to check for it
in the case statements, just let it get caught by the default: case.

Fixes: e3614bc9dc44 ("IB/ipoib: Add readout of statistics using ethtool")
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA: Convert CQ allocations to be under core responsibility

Ensure that CQ is allocated and freed by IB/core and not by drivers.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Gal Pressman <galpress@amazon.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Tested-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA: Clean destroy CQ in drivers do not return errors

Like all other destroy commands, .destroy_cq() call is not supposed
to fail. In all flows, the attempt to return earlier caused to memory
leaks.

This patch converts .destroy_cq() to do not return any errors.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Gal Pressman <galpress@amazon.com>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/nes: Avoid memory allocation during CQ destroy

The memory allocation call can fail and cause to early return
from nes_desotroy_cq() function. This situation will cause to
memory leak of struct nes_cq. Rewrite function to avoid memory
allocation.

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

RDMA: Move owner into struct ib_device_ops

This more closely follows how other subsytems work, with owner being a
member of the structure containing the function pointers.

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

RDMA: Move uverbs_abi_ver into struct ib_device_ops

No reason for every driver to emit code to set this, just make it part of
the driver's existing static const ops structure.

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

RDMA: Move driver_id into struct ib_device_ops

No reason for every driver to emit code to set this, just make it part of
the driver's existing static const ops structure.

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

rdma: Delete the ib_ucm module

This has been marked CONFIG_BROKEN for over a year now with no complaints.
Delete the whole thing for good.

The module provided the /dev/infiniband/ucmX interface.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

ucma: Convert ctx_idr to XArray

Signed-off-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

ucma: Convert multicast_idr to XArray

Signed-off-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/ucma: Use struct_size() helper

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Bugfix for filling the sge of srq

When user post recv a srq with multiple sges, the hardware will get the
last correct sge and count the sge numbers according to the specific
identifier with lkey. For example, when the driver fills the sges with
every wr less than the max sge that the user configured when creating srq,
the hardware will stop getting the sge according to the specific lkey in
the sge. However, it will always end with the first sge in the current
post srq recv interface implementation.

Fixes: c7bcb13442e1 ("RDMA/hns: Add SRQ support for hip08 kernel mode")
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: fix inverted logic of readl read and shift

A previous change incorrectly changed the inverted logic and logically
negated the readl rather than the shifted readl result. Fix this by
adding in missing parentheses around the expression that needs to be
logically negated.

Addresses-Coverity: ("Logically dead code")
Fixes: 669cefb654cb ("RDMA/hns: Remove jiffies operation in disable interrupt context")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/srp: Accept again source addresses that do not have a port number

The function srp_parse_in() is used both for parsing source address
specifications and for target address specifications. Target addresses
must have a port number. Having to specify a port number for source
addresses is inconvenient. Make sure that srp_parse_in() supports again
parsing addresses with no port number.

Cc: <stable@vger.kernel.org>
Fixes: c62adb7def71 ("IB/srp: Fix IPv6 address parsing")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/ipoib: implement ethtool .get_link() callback

Add support for reporting link state for ipoib net devices.

$ ip l set dev mlx4_ib0 up
$ ethtool mlx4_ib0 | grep Link
Link detected: yes
$ ip l set dev mlx4_ib0 down
$ ethtool mlx4_ib0 | grep Link
Link detected: no

Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

{IB,net}/mlx5: Constify rep ops functions pointers

Currently for every representor type and for every single vport,
representer function pointers copy is stored even though they don't
change from one to other vport.

Additionally priv data entry for the rep is not passed during
registration, but its copied. It is used (set and cleared) by the user
of the reps.

As we want to scale vports, to simplify and also to split constants
from data,

1. Rename mlx5_eswitch_rep_if to mlx5_eswitch_rep_ops as to match _ops
prefix with other standard netdev, ibdev ops.
2. Constify the IB and Ethernet rep ops structure.
3. Instead of storing copy of all rep function pointers, store copy
per eswitch rep type.
4. Split data and function pointers to mlx5_eswitch_rep_ops and
mlx5_eswitch_rep_data.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

{IB, net}/mlx5: No need to typecast from void* to mlx5_ib_dev*

Avoid typecasting from void* to mlx5_ib_dev* or mlx5e_rep_priv*
as it is not needed.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: E-Switch, Honor eswitch functions changed event cap

Whenever device supports eswitch functions changed event, honor
such device setting. Do not limit it to ECPF.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Vu Pham <vuhuong@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: E-Switch, Replace host_params event with functions_changed event

To support sriov on a E-Switch manager, num_vfs are queried
to the firmware whenever E-Switch manager is notified by
esw_functions_changed event.

Replace host_params event with esw_functions_changed event that reflects
more appropriate naming.

While at it, also correct num_vfs type from int to u16 as expected by
the function mlx5_esw_query_functions().

Signed-off-by: Vu Pham <vuhuong@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Introduce termination table bits

Termination table is a flow table with a termination flag. The flag
allows the firmware to assume that the the specified actions are the last
actions list. This assumption allows the FW to safely perform potential
looping logic (e.g. hairpin). Introduce the bits for this attribute.

Signed-off-by: Eli Britstein <elibr@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

net/mlx5: Add core dump register access HW bits

Add Firmware core dump registers and HW definitions.

Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

RDMA/hns: Bugfix for posting multiple srq work request

When the user submits more than 32 work request to a srq queue
at a time, it needs to find the corresponding number of entries
in the bitmap in the idx queue. However, the original lookup
function named ffs only processes 32 bits of the array element,
When the number of srq wqe issued exceeds 32, the ffs will only
process the lower 32 bits of the elements, it will not be able
to get the correct wqe index for srq wqe.

Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/uverbs: check for allocation failure in uapi_add_elm()

If the kzalloc() fails then we should return ERR_PTR(-ENOMEM).  In the
current code it's possible that the kzalloc() fails and the
radix_tree_insert() inserts the NULL pointer successfully and we return
the NULL "elm" pointer to the caller.  That results in a NULL pointer
dereference.

Fixes: 9ed3e5f44772 ("IB/uverbs: Build the specs into a radix tree at runtime")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Use struct_size() helper

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes, in particular in the
context in which this code is being used.

So, replace the following form:

sizeof(struct opa_port_status_rsp) + num_vls * sizeof(struct _vls_pctrs)

with:

struct_size(rsp, vls, num_vls)

and so on...

Also, notice that variable size is unnecessary, hence it is removed.

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/qib: Use struct_size() helper

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes, in particular in the
context in which this code is being used.

So, replace the following form:

sizeof(*pkt) + sizeof(pkt->addr[0])*n

with:

struct_size(pkt, addr, n)

Also, notice that variable size is unnecessary, hence it is removed.

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/rdmavt: Use struct_size() helper

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes, in particular in the
context in which this code is being used.

So, replace the following form:

sizeof(struct rvt_sge) * init_attr->cap.max_send_sge + sizeof(struct rvt_swqe)

with:

struct_size(swq, sg_list, init_attr->cap.max_send_sge)

and so on...

Also, notice that variable size is unnecessary, hence it is removed.

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/efa: Remove unused includes

Remove leftover includes that are no longer used from the driver.

Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/efa: Use rdma block iterator in chunk list creation

When creating the chunks list the rdma_for_each_block() iterator is used
in order to iterate over the payload in EFA_CHUNK_PAYLOAD_SIZE (device
defined) strides.

Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/efa: Remove unneeded admin commands abort flow

The admin commands abort flow is buggy (use-after-free) and not really
necessary as it is guaranteed that after ib_unregister_device() is called
there are no user verbs threads running in parallel, delete it.

Suggested-by: Jason Gunthorpe <jgg@ziepe.ca>
Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/efa: Use kvzalloc instead of kzalloc with fallback

Use kvzalloc which attempts to allocate a physically continuous buffer and
fallbacks to virtually continuous on failure instead of open coding it in
the driver.

The is_vmalloc_addr function is used to determine whether the buffer is
physically continuous or not (which determines direct vs indirect MR
registration mode).

Suggested-by: Jason Gunthorpe <jgg@ziepe.ca>
Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove extra brackets from an if

A recent patch to hfi1 left behind a checkpatch error.

Fixes: fb24ea52f78e ("drivers: Remove explicit invocations of mmiowb()")
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

net/mlx5: potential error pointer dereference in error handling

The error handling was a bit flipped around.  If the mlx5_create_flow_group()
function failed then it would have resulted in dereferencing "fg" when
it was an error pointer.

Fixes: 80f09dfc237f ("net/mlx5: Eswitch, enable RoCE loopback traffic")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

RDMA: Convert put_page() to put_user_page*()

For infiniband code that retains pages via get_user_pages*(), release
those pages via the new put_user_page(), or put_user_pages*(), instead of
put_page()

This is a tiny part of the second step of fixing the problem described in
[1]. The steps are:

1) Provide put_user_page*() routines, intended to be used for releasing
   pages that were pinned via get_user_pages*().

2) Convert all of the call sites for get_user_pages*(), to invoke
   put_user_page*(), instead of put_page(). This involves dozens of call
   sites, and will take some time.

3) After (2) is complete, use get_user_pages*() and put_user_page*() to
   implement tracking of these pages. This tracking will be separate from
   the existing struct page refcounting.

4) Use the tracking and identification of these pages, to implement
   special handling (especially in writeback paths) when the pages are
   backed by a filesystem. Again, [1] provides details as to why that is
   desirable.

[1] https://lwn.net/Articles/753027/ : "The Trouble with get_user_pages()"

Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Jason Gunthorpe <jgg@mellanox.com>
Tested-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove set but not used variables 'offset' and 'fspsn'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/infiniband/hw/hfi1/tid_rdma.c: In function tid_rdma_rcv_error:
drivers/infiniband/hw/hfi1/tid_rdma.c:2029:7: warning: variable offset set but not used [-Wunused-but-set-variable]
drivers/infiniband/hw/hfi1/tid_rdma.c: In function hfi1_rc_rcv_tid_rdma_ack:
drivers/infiniband/hw/hfi1/tid_rdma.c:4555:35: warning: variable fspsn set but not used [-Wunused-but-set-variable]

'offset' is never used since introduction in
commit d0d564a1caac ("IB/hfi1: Add functions to receive TID RDMA READ request")

'fspsn' is never used since introduciotn in
commit 9e93e967f7b4 ("IB/hfi1: Add a function to receive TID RDMA ACK packet")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Replace magic numbers with #defines

This patch makes the code more readable by removing magic numbers.

Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Remove jiffies operation in disable interrupt context

In some functions, the jiffies operation is unnecessary, and we can
control delay using mdelay and udelay functions only.  Especially, in
hns_roce_v1_clear_hem, the function calls spin_lock_irqsave, the context
disables interrupt, so we can not use jiffies and msleep functions.

Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Move spin_lock_irqsave to the correct place

When hip08 set gid, it will call spin_unlock_bh when send cmq.  if main.ko
call spin_lock_irqsave firstly, and the kernel is before commit
f71b74bca637 ("irq/softirqs: Use lockdep to assert IRQs are
disabled/enabled"), it will cause WARN_ON_ONCE because of calling
spin_unlock_bh in disable context.

In fact, the spin_lock_irqsave in main.ko is only used for hip06, and
should be placed in hns_roce_hw_v1.c. hns_roce_hw_v2.c uses its own
spin_unlock_bh and does not need main.ko manage spin_lock.

Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/hns: Update CQE specifications

According to hip08 UM, the maximum number of CQEs supported by each CQ is
4M.

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

RDMA/hns: Remove unnecessary print message in aeq

There is no need to print when communication is established, especially
while lots of qp used by application.

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

iw_cxgb4: Fix qpid leak

Add await in destroy_qp() so that all references to qp are dereferenced
and qp is freed in destroy_qp() itself.  This ensures freeing of all QPs
before invocation of dealloc_ucontext(), which prevents loss of in use
qpids stored in the ucontext.

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

RDMA/cxgb4: Don't expose DMA addresses

Change unconditional print of DMA address to be printed with special
printk format type specifier.

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

RDMA/cxgb4: Use sizeof() notation

Convert various sizeof call sites to be written in standard format
sizeof().

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

RDMA/cxgb3: Delete and properly mark unimplemented resize CQ function

Resize CQ implementation was guarded by undeclared "notyet" define while
cxgb3 was added to the kernel. Twelve years later, this call is still
unimplemented, so safely delete it and fix improper return error code when
.resize_cq() is not implemented.

Fixes: b038ced7b370 ("RDMA/cxgb3: Add driver for Chelsio T3 RNIC")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/cxgb3: Don't expose DMA addresses

DMA addresses like all other kernel addresses should be printed with
special %p* formatter. It is needed to allow control of exposure of such
information through a dedicated knob.

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

RDMA/cxgb3: Use sizeof() notation instead of plain sizeof

sizeof(a), sizeof a and sizeof (a) are all valid notations, but first is
more readable format recommended by checkpatch.pl.

Let's canonize it in cxgb3 drivers, so latter patches won't emit
checkpatch warnings. As part of this change, a redundant memset() was
removed.

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

RDMA/efa: Remove check that prevents destroy of resources in error flows

Drivers cannot check the udata for validity when doing destroy as there
will be no way to report this error back to the uverbs.

Since udata is new for destroy no driver should start to use it - instead
drivers should opt for the ioctl interface and define it in a way where it
cannot fail due to incorrect data.

Remove the checks on udata construction so EFA is consistent with
everything else.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/nes: Remove second wait queue initialization call

The same wait queue is initialized a couple of lines above.

Fixes: 3c2d774cad5b ("RDMA/nes: Add a driver for NetEffect RNICs")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/i40iw: Remove useless NULL checks

There is no need to check existence of structures to be destroyed.

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

RDMA/nes: Remove useless NULL checks

The destroy functions are always called with relevant structs, there is no
need to check their existence.

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

RDMA/core: Make ib_destroy_cq() void

Kernel destroy CQ flows can't fail and the returned value of
ib_destroy_cq() is not interested in those flows.

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

RDMA/ipoib: Remove check of destroy CQ

There are nothing to do from user side with knowledge that destroy CQ
fails.

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

rds: Don't check return value from destroy CQ

There is no value in checking ib_destroy_cq() result and skipping to clear
struct ic fields. This connection needs to be reinitialized anyway.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Return void from ib_device_check_mandatory()

The return value from ib_device_check_mandatory() is always 0 - change it
to be void.

Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx4: Delete unused func arg

The function argument virt_addr is not in use - delete it.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/umem: Move page_shift from ib_umem to ib_odp_umem

This value has always been set to PAGE_SHIFT in the core code, the only
thing that does differently was the ODP path. Move the value into the ODP
struct and still use it for ODP, but change all the non-ODP things to just
use PAGE_SHIFT/PAGE_SIZE/PAGE_MASK directly.

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