RDMA/core: Simplify how the port sysfs is created

Use the same technique as gid_attrs now uses to manage the port
sysfs. Bundle everything into three allocations and use a single
sysfs_create_groups() to build everything in one shot.

All the memory is always freed in the kobj release function, removing most
of the error unwinding.

The gid_attr technique and the hw_counters are very similar, merge the two
together and combine the sysfs_create_group() call for hw_counters with
the single sysfs group setup.

Link: https://lore.kernel.org/r/b688f3340694c59f7b44b1bde40e25559ef43cf3.1623427137.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Simplify how the gid_attrs sysfs is created

Instead of having an whole bunch of different allocations to create the
gid_attr kobjects reduce it to three, one for the kobj struct plus the
attributes, and one for the attribute list for each of the two
groups.

Move the freeing of all allocations to the release function.

Reorder the operations so all the allocations happen first then the
kobject & sysfs operations are last.

This removes the majority of the complicated error unwind since the
release function will always undo all the memory allocations. Freeing the
memory is also much simpler since there is a lot less of it.

Consolidate creating the "group of array indexes" pattern into one helper
function. Ensure kobject_del is used.

Link: https://lore.kernel.org/r/f4149d379db7178d37d11d75e3026bf550f818a1.1623427137.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Split gid_attrs related sysfs from add_port()

The gid_attrs directory is a dedicated kobj nested under the port,
construct/destruct it with its own pair of functions for
understandability. This is much more readable than having it weirdly
inlined out of order into the add_port() function.

Link: https://lore.kernel.org/r/1c9434111b6770a7aef0e644a88a16eee7e325b8.1623427137.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Split port and device counter sysfs attributes

This code creates a 'struct hw_stats_attribute' for each sysfs entry that
contains a naked 'struct attribute' inside.

It then proceeds to attach this same structure to a 'struct device' kobj
and a 'struct ib_port' kobj. However, this violates the typing
requirements.  'struct device' requires the attribute to be a 'struct
device_attribute' and 'struct ib_port' requires the attribute to be
'struct port_attribute'.

This happens to work because the show/store function pointers in all three
structures happen to be at the same offset and happen to be nearly the
same signature. This means when container_of() was used to go between the
wrong two types it still managed to work.

However clang CFI detection notices that the function pointers have a
slightly different signature. As with show/store this was only working
because the device and port struct layouts happened to have the kobj at
the front.

Correct this by have two independent sets of data structures for the port
and device case. The two different attributes correctly include the
port/device_attribute struct and everything from there up is kept
split. The show/store function call chains start with device/port unique
functions that invoke a common show/store function pointer.

Link: https://lore.kernel.org/r/a8b3864b4e722aed3657512af6aa47dc3c5033be.1623427137.git.leonro@nvidia.com
Reported-by: Nathan Chancellor <nathan@kernel.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Replace the ib_port_data hw_stats pointers with a ib_port pointer

It is much saner to store a pointer to the kobject structure that contains
the cannonical stats pointer than to copy the stats pointers into a public
structure.

Future patches will require the sysfs pointer for other purposes.

Link: https://lore.kernel.org/r/f90551dfd296cde1cb507bbef27cca9891d19871.1623427137.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA: Split the alloc_hw_stats() ops to port and device variants

This is being used to implement both the port and device global stats,
which is causing some confusion in the drivers. For instance EFA and i40iw
both seem to be misusing the device stats.

Split it into two ops so drivers that don't support one or the other can
leave the op NULL'd, making the calling code a little simpler to
understand.

Link: https://lore.kernel.org/r/1955c154197b2a159adc2dc97266ddc74afe420c.1623427137.git.leonro@nvidia.com
Tested-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Disallow MR dereg and invalidate when bound

Check that an MR has no bound MWs before allowing a dereg or invalidate
operation.

Link: https://lore.kernel.org/r/20210608042552.33275-11-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Implement memory access through MWs

Add code to implement memory access through memory windows.

Link: https://lore.kernel.org/r/20210608042552.33275-10-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Implement invalidate MW operations

Implement invalidate MW and cleaned up invalidate MR operations.

Added code to perform remote invalidate for send with invalidate.  Added
code to perform local invalidation. Deleted some blank lines in rxe_loc.h.

Link: https://lore.kernel.org/r/20210608042552.33275-9-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Add support for bind MW work requests

Add support for bind MW work requests from user space.  Since rdma/core
does not support bind mw in ib_send_wr there is no way to support bind mw
in kernel space.

Added bind_mw local operation in rxe_req.c. Added bind_mw WR operation in
rxe_opcode.c. Added bind_mw WC in rxe_comp.c.  Added additional fields to
rxe_mw in rxe_verbs.h. Added rxe_do_dealloc_mw() subroutine to cleanup an
mw when rxe_dealloc_mw is called.  Added code to implement bind_mw
operation in rxe_mw.c

Link: https://lore.kernel.org/r/20210608042552.33275-8-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Move local ops to subroutine

Simplify rxe_requester() by moving the local operations to a subroutine.
Add an error return for illegal send WR opcode.  Moved next_index ahead of
rxe_run_task which fixed a small bug where work completions were delayed
until after the next wqe which was not the intended behavior.  Let errors
return their own WC status. Previously all errors were reported as
protection errors which was incorrect. Changed the return of errors from
rxe_do_local_ops() to err: which causes an immediate completion.  Without
this an error on a last WR may get lost. Changed fill_packet() to
finish_packet() which is more accurate.

Fixes: 8700e2e7c485 ("The software RoCE driver")
Link: https://lore.kernel.org/r/20210608042552.33275-7-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Replace WR_REG_MASK by WR_LOCAL_OP_MASK

Rxe has two mask bits WR_LOCAL_MASK and WR_REG_MASK with WR_REG_MASK used
to indicate any local operation and WR_LOCAL_MASK unused. This patch
replaces both of these with one mask bit WR_LOCAL_OP_MASK which is
clearer.

Link: https://lore.kernel.org/r/20210608042552.33275-6-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Add ib_alloc_mw and ib_dealloc_mw verbs

Add ib_alloc_mw and ib_dealloc_mw verbs APIs.

Added new file rxe_mw.c focused on MWs. Changed the 8 bit random key
generator. Added a cleanup routine for MWs. Added verbs routines to
ib_device_ops.

Link: https://lore.kernel.org/r/20210608042552.33275-5-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Enable MW object pool

Currently the rxe driver has a rxe_mw struct object but nothing about
memory windows is enabled. This patch turns on memory windows and some
minor cleanup.

Set device attribute in rxe.c so max_mw = MAX_MW.  Change parameters in
rxe_param.h so that MAX_MW is the same as MAX_MR.  Reduce the number of
MRs and MWs to 4K from 256K.  Add device capability bits for 2a and 2b
memory windows.  Removed RXE_MR_TYPE_MW from the rxe_mr_type enum.

Link: https://lore.kernel.org/r/20210608042552.33275-4-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Return errors for add index and key

Modify rxe_add_index() and rxe_add_key() to return an error if the index
or key is aleady present in the pool.  Currently they print a warning and
silently fail with bad consequences to the caller.

Link: https://lore.kernel.org/r/20210608042552.33275-3-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Add bind MW fields to rxe_send_wr

Add fields to struct rxe_send_wr in rdma_user_rxe.h to support bind MW
work requests

Link: https://lore.kernel.org/r/20210608042552.33275-2-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix qp reference counting for atomic ops

Currently the rdma_rxe driver attempts to protect atomic responder
resources by taking a reference to the qp which is only freed when the
resource is recycled for a new read or atomic operation. This means that
in normal circumstances there is almost always an extra qp reference once
an atomic operation has been executed which prevents cleaning up the qp
and associated pd and cqs when the qp is destroyed.

This patch removes the call to rxe_add_ref() in send_atomic_ack() and the
call to rxe_drop_ref() in free_rd_atomic_resource(). If the qp is
destroyed while a peer is retrying an atomic op it will cause the
operation to fail which is acceptable.

Link: https://lore.kernel.org/r/20210604230558.4812-1-rpearsonhpe@gmail.com
Reported-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Fixes: 86af61764151 ("IB/rxe: remove unnecessary skb_clone")
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Support getting max QP number from firmware

All functions of HIP09's ROCEE share on-chip resources for all QPs, the
driver needs configure the resource index and number for each function
during the init stage.

Link: https://lore.kernel.org/r/1622541427-42193-1-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Don't add slave port to unaffiliated list

The mlx5_ib_bind_slave_port() doesn't remove multiport device from the
unaffiliated list, but mlx5_ib_unbind_slave_port() did it. This unbalanced
flow caused to the situation where mlx5_ib_unaffiliated_port_list was
changed during iteration.

Fixes: 32f69e4be269 ("{net, IB}/mlx5: Manage port association for multiport RoCE")
Link: https://lore.kernel.org/r/2726e6603b1e6ecfe76aa5a12a063af72173bcf7.1622477058.git.leonro@nvidia.com
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Store PBL info address a pointer type

The level1 PBL info address is stored as u64.  This requires casting
through a uinptr_t before used as a pointer type.

And this leads to sparse warning such as this when uinptr_t is missing:

drivers/infiniband/hw/irdma/hw.c: In function 'irdma_destroy_virt_aeq':
drivers/infiniband/hw/irdma/hw.c:579:23: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
  579 |  dma_addr_t *pg_arr = (dma_addr_t *)aeq->palloc.level1.addr;

This can be fixed using an intermediate uintptr_t, but rather it is better
to fix the structure irdm_pble_info to store the address as u64* and the
VA it is assigned in irdma_chunk as a void*. This greatly reduces the
casting on this address.

Fixes: 44d9e52977a1 ("RDMA/irdma: Implement device initialization definitions")
Link: https://lore.kernel.org/r/20210609234924.938-1-shiraz.saleem@intel.com
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Remove dgid from the cm_id_priv av

It turns out this is only being used to store the LID for SIDR mode to
search the RB tree for request de-duplication. Store the LID value
directly and don't pretend it is a GID.

Link: https://lore.kernel.org/r/2e7c87b6f662c90c642fc1838e363ad3e6ef14a4.1623236345.git.leonro@nvidia.com
Reviewed-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Use list_last_entry/list_first_entry

Use list_last_entry and list_first_entry instead of using prev and next
pointers.

Link: https://lore.kernel.org/r/20210608211415.680-1-shiraz.saleem@intel.com
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Use list_move instead of list_del/list_add

Using list_move() instead of list_del() + list_add().

Link: https://lore.kernel.org/r/20210608031041.2820429-1-libaokun1@huawei.com
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Use refcount_t instead of atomic_t on refcount of ib_uverbs_device

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-8-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/ipoib: Use refcount_t instead of atomic_t for reference counting

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-13-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cxgb4: Use refcount_t instead of atomic_t for reference counting

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-12-git-send-email-liweihang@huawei.com
Cc: Potnuri Bharat Teja <bharat@chelsio.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Use refcount_t instead of atomic_t for QP reference counting

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-11-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Use refcount_t instead of atomic_t for SRQ reference counting

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-10-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Use refcount_t instead of atomic_t for CQ reference counting

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-9-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Use refcount_t instead of atomic_t on refcount of mcast_port

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-6-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Use refcount_t instead of atomic_t on refcount of mcast_member

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-5-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Remove refcount from struct ib_mad_snoop_private

The member is never used, delete it.

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

RDMA/core: Use refcount_t instead of atomic_t on refcount of iwpm_admin_data

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks. Increase refcount_t from 0 to 1 is
regarded as there is a risk about use-after-free. So it should be set to 1
directly during initialization.

Link: https://lore.kernel.org/r/1622194663-2383-3-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Use refcount_t instead of atomic_t on refcount of iwcm_id_private

The refcount_t API will WARN on underflow and overflow of a reference
counter, and avoid use-after-free risks.

Link: https://lore.kernel.org/r/1622194663-2383-2-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Fix return error sign from irdma_modify_qp

There is a typo in the returned error code sign from irdma_modify_qp()
when the attr_mask is not supported - Fix it.

Fixes: b48c24c2d710 ("RDMA/irdma: Implement device supported verb APIs")
Link: https://lore.kernel.org/r/20210607221543.254144-1-kamalheib1@gmail.com
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Fix spelling mistake "Allocal" -> "Allocate"

There is a spelling mistake in a literal string. Fix it.

Link: https://lore.kernel.org/r/20210607113345.82206-1-colin.king@canonical.com
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: remove redundant initialization of variable val

The variable val is being initialized with a value that is never
read, it is being updated later on. The assignment is redundant and
can be removed.

Link: https://lore.kernel.org/r/20210605131347.26293-1-colin.king@canonical.com
Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: remove extraneous indentation on a statement

A single statement is indented one level too deeply, clean up the
code by removing the extraneous tab.

Link: https://lore.kernel.org/r/20210605130400.25987-1-colin.king@canonical.com
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Fix issues with u8 left shift operation

The shifting of the u8 integer info->map[i] the left will be promoted
to a 32 bit signed int and then sign-extended to a u64. In the event
that the top bit of the u8 is set then all then all the upper 32 bits
of the u64 end up as also being set because of the sign-extension.
Fix this by casting the u8 values to a u64 before the left shift. This

Link: https://lore.kernel.org/r/20210605122059.25105-1-colin.king@canonical.com
Addresses-Coverity: ("Unitentional integer overflow / bad shift operation")
Fixes: 3f49d6842569 ("RDMA/irdma: Implement HW Admin Queue OPs")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cxgb4: Fix missing error code in create_qp()

The error code is missing in this code scenario so 0 will be returned. Add
the error code '-EINVAL' to the return value 'ret'.

Eliminates the follow smatch warning:

drivers/infiniband/hw/cxgb4/qp.c:298 create_qp() warn: missing error code 'ret'.

Link: https://lore.kernel.org/r/1622545669-20625-1-git-send-email-jiapeng.chong@linux.alibaba.com
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Update maintainers list

Updated the maintainers list and removed non-active members.

Link: https://lore.kernel.org/r/20210603131534.982257-3-devesh.sharma@broadcom.com
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Enable global atomic ops if platform supports

Enabling Atomic operations for Gen P5 devices if the underlying platform
supports global atomic ops.

Link: https://lore.kernel.org/r/20210603131534.982257-2-devesh.sharma@broadcom.com
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix failure during driver load

To avoid the following failure when trying to load the rdma_rxe module
while IPv6 is disabled, add a check for EAFNOSUPPORT and ignore the
failure, also delete the needless debug print from rxe_setup_udp_tunnel().

$ modprobe rdma_rxe
modprobe: ERROR: could not insert 'rdma_rxe': Operation not permitted

Fixes: dfdd6158ca2c ("IB/rxe: Fix kernel panic in udp_setup_tunnel")
Link: https://lore.kernel.org/r/20210603090112.36341-1-kamalheib1@gmail.com
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Protext kernel index from user space

In order to prevent user space from modifying the index that belongs to
the kernel for shared queues let the kernel use a local copy of the index
and copy any new values of that index to the shared rxe_queue_bus struct.

This adds more switch statements which decreases the performance of the
queue API. Move the type into the parameter list for these functions so
that the compiler can optimize out the switch statements when the explicit
type is known. Modify all the calls in the driver on performance paths to
pass in the explicit queue type.

Link: https://lore.kernel.org/r/20210527194748.662636-4-rpearsonhpe@gmail.com
Link: https://lore.kernel.org/linux-rdma/20210526165239.GP1002214@@nvidia.com/
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Protect user space index loads/stores

Modify the queue APIs to protect all user space index loads with
smp_load_acquire() and all user space index stores with
smp_store_release(). Base this on the types of the queues which can be one
of ..KERNEL, ..FROM_USER, ..TO_USER. Kernel space indices are protected by
locks which also provide memory barriers.

Link: https://lore.kernel.org/r/20210527194748.662636-3-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Add a type flag to rxe_queue structs

To create optimal code only want to use smp_load_acquire() and
smp_store_release() for user indices in rxe_queue APIs since kernel
indices are protected by locks which also act as memory barriers. By
adding a type to the queues we can determine which indices need to be
protected.

Link: https://lore.kernel.org/r/20210527194748.662636-2-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Merge branch 'irdma' into rdma.git for-next

Shiraz Saleem says:

====================
Add Intel Ethernet Protocol Driver for RDMA (irdma)

The following patch series introduces a unified Intel Ethernet Protocol
Driver for RDMA (irdma) for the X722 iWARP device and a new E810 device
which supports iWARP and RoCEv2. The irdma module replaces the legacy
i40iw module for X722 and extends the ABI already defined for i40iw. It is
backward compatible with legacy X722 rdma-core provider (libi40iw).

X722 and E810 are PCI network devices that are RDMA capable. The RDMA
block of this parent device is represented via an auxiliary device
exported to 'irdma' using the core auxiliary bus infrastructure recently
added for 5.11 kernel.  The parent PCI netdev drivers 'i40e' and 'ice'
register auxiliary RDMA devices with private data/ops encapsulated that
bind to auxiliary drivers registered in irdma module.

Currently, default is RoCEv2 for E810. Runtime support for protocol switch
to iWARP will be made available via devlink in a future patch.
====================

Link: https://lore.kernel.org/r/20210602205138.889-1-shiraz.saleem@intel.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
* branch 'irdma':
  RDMA/irdma: Update MAINTAINERS file
  RDMA/irdma: Add irdma Kconfig/Makefile and remove i40iw
  RDMA/irdma: Add ABI definitions
  RDMA/irdma: Add dynamic tracing for CM
  RDMA/irdma: Add miscellaneous utility definitions
  RDMA/irdma: Add user/kernel shared libraries
  RDMA/irdma: Add RoCEv2 UD OP support
  RDMA/irdma: Implement device supported verb APIs
  RDMA/irdma: Add PBLE resource manager
  RDMA/irdma: Add connection manager
  RDMA/irdma: Add QoS definitions
  RDMA/irdma: Add privileged UDA queue implementation
  RDMA/irdma: Add HMC backing store setup functions
  RDMA/irdma: Implement HW Admin Queue OPs
  RDMA/irdma: Implement device initialization definitions
  RDMA/irdma: Register auxiliary driver and implement private channel OPs
  i40e: Register auxiliary devices to provide RDMA
  i40e: Prep i40e header for aux bus conversion
  ice: Register auxiliary device to provide RDMA
  ice: Implement iidc operations
  ice: Initialize RDMA support
  iidc: Introduce iidc.h
  i40e: Replace one-element array with flexible-array member

RDMA/irdma: Update MAINTAINERS file

Add maintainer entry for irdma driver.

Link: https://lore.kernel.org/r/20210602205138.889-17-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add irdma Kconfig/Makefile and remove i40iw

Add Kconfig and Makefile to build irdma driver.

Remove i40iw driver and add an alias in irdma.

Remove legacy exported symbols i40e_register_client
and i40e_unregister_client from i40e as they are no
longer used.

irdma is the replacement driver that supports X722.

Link: https://lore.kernel.org/r/20210602205138.889-16-shiraz.saleem@intel.com
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add ABI definitions

Add ABI definitions for irdma.

Link: https://lore.kernel.org/r/20210602205138.889-15-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add dynamic tracing for CM

Add dynamic tracing functionality to debug connection
management issues.

Link: https://lore.kernel.org/r/20210602205138.889-14-shiraz.saleem@intel.com
Signed-off-by: "Michael J. Ruhl" <michael.j.ruhl@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add miscellaneous utility definitions

Add miscellaneous utility functions and headers.

Link: https://lore.kernel.org/r/20210602205138.889-13-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add user/kernel shared libraries

Building the WQE descriptors for different verb
operations are similar in kernel and user-space.
Add these shared libraries.

Link: https://lore.kernel.org/r/20210602205138.889-12-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add RoCEv2 UD OP support

Add the header, data structures and functions
to populate the WQE descriptors and issue the
Control QP commands that support RoCEv2 UD operations.

Link: https://lore.kernel.org/r/20210602205138.889-11-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Implement device supported verb APIs

Implement device supported verb APIs. The supported APIs
vary based on the underlying transport the ibdev is
registered as (i.e. iWARP or RoCEv2).

Link: https://lore.kernel.org/r/20210602205138.889-10-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add PBLE resource manager

Implement a Physical Buffer List Entry (PBLE) resource manager
to manage a pool of PBLE HMC resource objects.

Link: https://lore.kernel.org/r/20210602205138.889-9-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add connection manager

Add connection management (CM) implementation for
iWARP including accept, reject, connect, create_listen,
destroy_listen and CM utility functions

Link: https://lore.kernel.org/r/20210602205138.889-8-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add QoS definitions

Add definitions for managing the RDMA HW work scheduler (WS) tree.

A WS node is created via a control QP operation with the bandwidth
allocation, arbitration scheme, and traffic class of the QP specified.
The Qset handle returned associates the QoS parameters for the QP.
The Qset is registered with the LAN and an equivalent node is created
in the LAN packet scheduler tree.

Link: https://lore.kernel.org/r/20210602205138.889-7-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add privileged UDA queue implementation

Implement privileged UDA queues to handle iWARP connection
packets and receive exceptions.

Link: https://lore.kernel.org/r/20210602205138.889-6-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Add HMC backing store setup functions

HW uses host memory as a backing store for a number of
protocol context objects and queue state tracking.
The Host Memory Cache (HMC) is a component responsible for
managing these objects stored in host memory.

Add the functions and data structures to manage the allocation
of backing pages used by the HMC for the various objects

Link: https://lore.kernel.org/r/20210602205138.889-5-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Implement HW Admin Queue OPs

The driver posts privileged commands to the HW
Admin Queue (Control QP or CQP) to request administrative
actions from the HW. Implement create/destroy of CQP
and the supporting functions, data structures and headers
to handle the different CQP commands

Link: https://lore.kernel.org/r/20210602205138.889-4-shiraz.saleem@intel.com
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Implement device initialization definitions

Implement device initialization routines, interrupt set-up,
and allocate object bit-map tracking structures.
Also, add device specific attributes and register definitions.

Link: https://lore.kernel.org/r/20210602205138.889-3-shiraz.saleem@intel.com
[flexible array transformation]
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Register auxiliary driver and implement private channel OPs

Register auxiliary drivers which can attach to auxiliary RDMA
devices from Intel PCI netdev drivers i40e and ice. Implement the private
channel ops, and register net notifiers.

Link: https://lore.kernel.org/r/20210602205138.889-2-shiraz.saleem@intel.com
[flexible array transformation]
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Protect cm_dev, cm_ports and mad_agent with kref and lock

During cm_dev deregistration in cm_remove_one(), the cm_device and
cm_ports will be freed, after that they should not be accessed. The
mad_agent needs to be protected as well.

This patch adds a cm_device kref to protect cm_dev and cm_ports, and a
mad_agent_lock spinlock to protect mad_agent.

Link: https://lore.kernel.org/r/501ba7a2ff203dccd0e6755d3f93329772adce52.1622629024.git.leonro@nvidia.com
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Improve the calling of cm_init_av_for_lap and cm_init_av_by_path

The cm_init_av_for_lap() and cm_init_av_by_path() function calls have the
following issues:

1. Both of them might sleep and should not be called under spinlock.
2. The access of cm_id_priv->av should be under cm_id_priv->lock, which
   means it can't be initialized directly.

This patch splits the calling of 2 functions into two parts: first one
initializes an AV outside of the spinlock, the second one copies AV to
cm_id_priv->av under spinlock.

Fixes: e1444b5a163e ("IB/cm: Fix automatic path migration support")
Link: https://lore.kernel.org/r/038fb8ad932869b4548b0c7708cab7f76af06f18.1622629024.git.leonro@nvidia.com
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Simplify ib_cancel_mad() and ib_modify_mad() calls

The mad_agent parameter is redundant since the struct ib_mad_send_buf
already has a pointer of it.

Link: https://lore.kernel.org/r/0987c784b25f7bfa72f78691f50cff066de587e1.1622629024.git.leonro@nvidia.com
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Revert "IB/cm: Mark stale CM id's whenever the mad agent was unregistered"

This reverts commit 9db0ff53cb9b43ed75bacd42a89c1a0ab048b2b0, which wasn't
a full fix and still causes to the following panic:

panic @ time 1605623870.843, thread 0xfffffeb63b552000: vm_fault_lookup: fault on nofault entry, addr: 0xfffffe811a94e000
    time = 1605623870
    cpuid = 9, TSC = 0xb7937acc1b6
    Panic occurred in module kernel loaded at 0xffffffff80200000:Stack: --------------------------------------------------
    kernel:vm_fault+0x19da
    kernel:vm_fault_trap+0x6e
    kernel:trap_pfault+0x1f1
    kernel:trap+0x31e
    kernel:cm_destroy_id+0x38c
    kernel:rdma_destroy_id+0x127
    kernel:sdp_shutdown_task+0x3ae
    kernel:taskqueue_run_locked+0x10b
    kernel:taskqueue_thread_loop+0x87
    kernel:fork_exit+0x83

Link: https://lore.kernel.org/r/4346449a7cdacc7a4eedc89cb1b42d8434ec9814.1622629024.git.leonro@nvidia.com
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Tidy remaining cm_msg free paths

Now that all the free paths are explicit cm_free_msg() will only be called
for msgs's allocated with cm_alloc_msg(), so we can assume the context is
set. Place it after the allocation function it is paired with for clarity.

Also remove a bogus NULL assignment in one place after a cancel. This does
nothing other than disable completions to become events, but changing the
state already did that.

Link: https://lore.kernel.org/r/082fd3552be0d1a2c19b1c4cefb5f3f0e3e68e82.1622629024.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Call the correct message free functions in cm_send_handler()

There are now three destroy functions for the cm_msg, and all places
except the general send completion handler use the correct function.

Fix cm_send_handler() to detect which kind of message is being completed
and destroy it using the correct function with the correct locking.

Link: https://lore.kernel.org/r/62a507195b8db85bb11228d0c6e7fa944204bf12.1622629024.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Split cm_alloc_msg()

This is being used with two quite different flows, one attaches the
message to the priv and the other does not.

Ensure the message attach is consistently done under the spinlock and
ensure that the free on error always detaches the message from the
cm_id_priv, also always under lock.

This makes read/write to the cm_id_priv->msg consistently locked and
consistently NULL'd when the message is freed, even in all error paths.

Link: https://lore.kernel.org/r/f692b8c89eecb34fd82244f317e478bea6c97688.1622629024.git.leonro@nvidia.com
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Pair cm_alloc_response_msg() with a cm_free_response_msg()

This is not a functional change, but it helps make the purpose of all the
cm_free_msg() calls clearer. In this case a response msg has a NULL
context[0], and is never placed in cm_id_priv->msg.

Link: https://lore.kernel.org/r/5cd53163be7df0a94f0d4ef7294546bc674fb74a.1622629024.git.leonro@nvidia.com
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Sanitize WQ state received from the userspace

The mlx4 and mlx5 implemented differently the WQ input checks.  Instead of
duplicating mlx4 logic in the mlx5, let's prepare the input in the central
place.

The mlx5 implementation didn't check for validity of state input.  It is
not real bug because our FW checked that, but still worth to fix.

Fixes: f213c0527210 ("IB/uverbs: Add WQ support")
Link: https://lore.kernel.org/r/ac41ad6a81b095b1a8ad453dcf62cf8d3c5da779.1621413310.git.leonro@nvidia.com
Reported-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Avoid Wtautological-constant-out-of-range-compare

drivers/infiniband/ulp/rtrs/rtrs-clt.c:1786:19: warning: result of comparison of
constant 'MAX_SESS_QUEUE_DEPTH' (65536) with expression of type 'u16'
(aka 'unsigned short') is always false [-Wtautological-constant-out-of-range-compare]

To fix it, limit MAX_SESS_QUEUE_DEPTH to u16 max, which is 65535, and
drop the check in rtrs-clt, as it's the type u16 max.

Link: https://lore.kernel.org/r/20210531122835.58329-1-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

i40e: Register auxiliary devices to provide RDMA

Convert i40e to use the auxiliary bus infrastructure to export
the RDMA functionality of the device to the RDMA driver.
Register i40e client auxiliary RDMA device on the auxiliary bus per
PCIe device function for the new auxiliary rdma driver (irdma) to
attach to.

The global i40e_register_client and i40e_unregister_client symbols
will be obsoleted once irdma replaces i40iw in the kernel
for the X722 device.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>

i40e: Prep i40e header for aux bus conversion

Add the definitions to the i40e client header file in
preparation to convert i40e to use the new auxiliary bus
infrastructure. This header is shared between the 'i40e'
Intel networking driver providing RDMA support and the
'irdma' driver.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>

ice: Register auxiliary device to provide RDMA

Register ice client auxiliary RDMA device on the auxiliary bus per
PCIe device function for the auxiliary driver (irdma) to attach to.
It allows to realize a single RDMA driver (irdma) capable of working with
multiple netdev drivers over multi-generation Intel HW supporting RDMA.
There is no load ordering dependencies between ice and irdma.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>

ice: Implement iidc operations

Add implementations for supporting iidc operations for device operation
such as allocation of resources and event notifications.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>

ice: Initialize RDMA support

Probe the device's capabilities to see if it supports RDMA. If so, allocate
and reserve resources to support its operation; populate structures with
initial values.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>

iidc: Introduce iidc.h

Introduce a shared header file used by the 'ice' Intel networking driver
providing RDMA support and the 'irdma' driver to provide a private
interface.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>

RDMA/rtrs-clt: Fix memory leak of not-freed sess->stats and stats->pcpu_stats

sess->stats and sess->stats->pcpu_stats objects are freed
when sysfs entry is removed. If something wrong happens and
session is closed before sysfs entry is created,
sess->stats and sess->stats->pcpu_stats objects are not freed.

This patch adds freeing of them at three places:
1. When client uses wrong address and session creation fails.
2. When client fails to create a sysfs entry.
3. When client adds wrong address via sysfs add_path.

Fixes: 215378b838df0 ("RDMA/rtrs: client: sysfs interface functions")
Link: https://lore.kernel.org/r/20210528113018.52290-21-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-clt: Check if the queue_depth has changed during a reconnection

The queue_depth is a module parameter for rtrs_server. It is used on the
client side to determing the queue_depth of the request queue for the RNBD
virtual block device.

During a reconnection event for an already mapped device, in case the
rtrs_server module queue_depth has changed, fail the reconnect attempt.

Also stop further auto reconnection attempts. A manual reconnect via
sysfs has to be triggerred.

Fixes: 6a98d71daea18 ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/20210528113018.52290-20-jinpu.wang@ionos.com
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Fix memory leak when having multiple sessions

Gioh notice memory leak below
unreferenced object 0xffff8880acda2000 (size 2048):
  comm "kworker/4:1", pid 77, jiffies 4295062871 (age 1270.730s)
  hex dump (first 32 bytes):
    00 20 da ac 80 88 ff ff 00 20 da ac 80 88 ff ff  . ....... ......
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<00000000e85d85b5>] rtrs_srv_rdma_cm_handler+0x8e5/0xa90 [rtrs_server]
    [<00000000e31a988a>] cma_ib_req_handler+0xdc5/0x2b50 [rdma_cm]
    [<000000000eb02c5b>] cm_process_work+0x2d/0x100 [ib_cm]
    [<00000000e1650ca9>] cm_req_handler+0x11bc/0x1c40 [ib_cm]
    [<000000009c28818b>] cm_work_handler+0xe65/0x3cf2 [ib_cm]
    [<000000002b53eaa1>] process_one_work+0x4bc/0x980
    [<00000000da3499fb>] worker_thread+0x78/0x5c0
    [<00000000167127a4>] kthread+0x191/0x1e0
    [<0000000060802104>] ret_from_fork+0x3a/0x50
unreferenced object 0xffff88806d595d90 (size 8):
  comm "kworker/4:1H", pid 131, jiffies 4295062972 (age 1269.720s)
  hex dump (first 8 bytes):
    62 6c 61 00 6b 6b 6b a5                          bla.kkk.
  backtrace:
    [<000000004447d253>] kstrdup+0x2e/0x60
    [<0000000047259793>] kobject_set_name_vargs+0x2f/0xb0
    [<00000000c2ee3bc8>] dev_set_name+0xab/0xe0
    [<000000002b6bdfb1>] rtrs_srv_create_sess_files+0x260/0x290 [rtrs_server]
    [<0000000075d87bd7>] rtrs_srv_info_req_done+0x71b/0x960 [rtrs_server]
    [<00000000ccdf1bb5>] __ib_process_cq+0x94/0x100 [ib_core]
    [<00000000cbcb60cb>] ib_cq_poll_work+0x32/0xc0 [ib_core]
    [<000000002b53eaa1>] process_one_work+0x4bc/0x980
    [<00000000da3499fb>] worker_thread+0x78/0x5c0
    [<00000000167127a4>] kthread+0x191/0x1e0
    [<0000000060802104>] ret_from_fork+0x3a/0x50
unreferenced object 0xffff88806d6bb100 (size 256):
  comm "kworker/4:1H", pid 131, jiffies 4295062972 (age 1269.720s)
  hex dump (first 32 bytes):
    00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00  .....N..........
    ff ff ff ff ff ff ff ff 00 59 4d 86 ff ff ff ff  .........YM.....
  backtrace:
    [<00000000a18a11e4>] device_add+0x74d/0xa00
    [<00000000a915b95f>] rtrs_srv_create_sess_files.cold+0x49/0x1fe [rtrs_server]
    [<0000000075d87bd7>] rtrs_srv_info_req_done+0x71b/0x960 [rtrs_server]
    [<00000000ccdf1bb5>] __ib_process_cq+0x94/0x100 [ib_core]
    [<00000000cbcb60cb>] ib_cq_poll_work+0x32/0xc0 [ib_core]
    [<000000002b53eaa1>] process_one_work+0x4bc/0x980
    [<00000000da3499fb>] worker_thread+0x78/0x5c0
    [<00000000167127a4>] kthread+0x191/0x1e0
    [<0000000060802104>] ret_from_fork+0x3a/0x50

The problem is we increase device refcount by get_device in process_info_req
for each path, but only does put_deice for last path, which lead to
memory leak.

To fix it, it also calls put_device when dev_ref is not 0.

Fixes: e2853c49477d1 ("RDMA/rtrs-srv-sysfs: fix missing put_device")
Link: https://lore.kernel.org/r/20210528113018.52290-19-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Fix memory leak of unfreed rtrs_srv_stats object

When closing a session, currently the rtrs_srv_stats object in the
closing session is freed by kobject release. But if it failed
to create a session by various reasons, it must free the rtrs_srv_stats
object directly because kobject is not created yet.

This problem is found by kmemleak as below:

1. One client machine maps /dev/nullb0 with session name 'bla':
root@test1:~# echo "sessname=bla path=ip:192.168.122.190 \
device_path=/dev/nullb0" > /sys/devices/virtual/rnbd-client/ctl/map_device

2. Another machine failed to create a session with the same name 'bla':
root@test2:~# echo "sessname=bla path=ip:192.168.122.190 \
device_path=/dev/nullb1" > /sys/devices/virtual/rnbd-client/ctl/map_device
-bash: echo: write error: Connection reset by peer

3. The kmemleak on server machine reported an error:
unreferenced object 0xffff888033cdc800 (size 128):
  comm "kworker/2:1", pid 83, jiffies 4295086585 (age 2508.680s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<00000000a72903b2>] __alloc_sess+0x1d4/0x1250 [rtrs_server]
    [<00000000d1e5321e>] rtrs_srv_rdma_cm_handler+0xc31/0xde0 [rtrs_server]
    [<00000000bb2f6e7e>] cma_ib_req_handler+0xdc5/0x2b50 [rdma_cm]
    [<00000000e896235d>] cm_process_work+0x2d/0x100 [ib_cm]
    [<00000000b6866c5f>] cm_req_handler+0x11bc/0x1c40 [ib_cm]
    [<000000005f5dd9aa>] cm_work_handler+0xe65/0x3cf2 [ib_cm]
    [<00000000610151e7>] process_one_work+0x4bc/0x980
    [<00000000541e0f77>] worker_thread+0x78/0x5c0
    [<00000000423898ca>] kthread+0x191/0x1e0
    [<000000005a24b239>] ret_from_fork+0x3a/0x50

Fixes: 39c2d639ca183 ("RDMA/rtrs-srv: Set .release function for rtrs srv device during device init")
Link: https://lore.kernel.org/r/20210528113018.52290-18-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Duplicated session name is not allowed

If two clients try to use the same session name, rtrs-server generates a
kernel error that it failed to create the sysfs because the filename
is duplicated.

This patch adds code to check if there already exists the same session
name with the different UUID. If a client tries to add more session,
it sends the UUID and the session name. Therefore it is ok if there is
already same session name with the same UUID. The rtrs-server must fail
only-if there is the same session name with the different UUID.

Link: https://lore.kernel.org/r/20210528113018.52290-17-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Aleksei Marov <aleksei.marov@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Do not reset hb_missed_max after re-connection

When re-connecting, it resets hb_missed_max to 0.
Before the first re-connecting, client will trigger re-connection
when it gets hb-ack more than 5 times. But after the first
re-connecting, clients will do re-connection whenever it does
not get hb-ack because hb_missed_max is 0.

There is no need to reset hb_missed_max when re-connecting.
hb_missed_max should be kept until closing the session.

Fixes: c0894b3ea69d3 ("RDMA/rtrs: core: lib functions shared between client and server modules")
Link: https://lore.kernel.org/r/20210528113018.52290-16-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: convert scnprintf to sysfs_emit

Link: https://lore.kernel.org/r/20210528113018.52290-15-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Replace atomic_t with percpu_ref for ids_inflight

ids_inflight is used to track the inflight IOs. But the use of atomic_t
variable can cause performance drops and can also become a performance
bottleneck.

This commit replaces the use of atomic_t with a percpu_ref structure. The
advantage it offers is, it doesn't check if the reference has fallen to 0,
until the user explicitly signals it to; and that is done by the
percpu_ref_kill() function call. After that, the percpu_ref structure
behaves like an atomic_t and for every put call, checks whether the
reference has fallen to 0 or not.

rtrs_srv_stats_rdma_to_str shows the count of ids_inflight as 0
for user-mode tools not to be confused.

Fixes: 9cb837480424e ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20210528113018.52290-14-jinpu.wang@ionos.com
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-clt: Check state of the rtrs_clt_sess before reading its stats

When get_next_path_min_inflight is called to select the next path, it
iterates over the list of available rtrs_clt_sess (paths). It then reads
the number of inflight IOs for that path to select one which has the least
inflight IO.

But it may so happen that rtrs_clt_sess (path) is no longer in the
connected state because closing or error recovery paths can change the status
of the rtrs_clt_Sess.

For example, the client sent the heart-beat and did not get the
response, it would change the session status and stop IO processing.
The added checking of this patch can prevent accessing the broken path
and generating duplicated error messages.

It is ok if the status is changed after checking the status because
the error recovery path does not free memory and only tries to
reconnection. And also it is ok if the session is closed after checking
the status because closing the session changes the session status and
flush all IO beforing free memory. If the session is being accessed for
IO processing, the closing session will wait.

Fixes: 6a98d71daea18 ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/20210528113018.52290-13-jinpu.wang@ionos.com
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Reviewed-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-clt: Remove redundant 'break'

It is duplicated with the very next line

Link: https://lore.kernel.org/r/20210528113018.52290-12-jinpu.wang@ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Kill __rtrs_srv_change_state

No need since the only user is rtrs_srv_change_state.

Link: https://lore.kernel.org/r/20210528113018.52290-11-jinpu.wang@ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-clt: Kill rtrs_clt_disconnect_from_sysfs

The function is just a wrapper of rtrs_clt_close_conns, let's call
rtrs_clt_close_conns directly.

Link: https://lore.kernel.org/r/20210528113018.52290-10-jinpu.wang@ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-clt: Kill rtrs_clt_{start,stop}_hb

The two wrappers are not needed since we can call rtrs_{start,stop}_hb
directly.

Link: https://lore.kernel.org/r/20210528113018.52290-9-jinpu.wang@ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Use strscpy instead of strlcpy

During checkpatch analyzing the following warning message was found:
  WARNING:STRLCPY: Prefer strscpy over strlcpy - see:
  https://lore.kernel.org/r/CAHk-=wgfRnXz0W3D37d01q3JFkr_i_uTL=V6A6G1oUZcprmknw@mail.gmail.com/
Fix it by using strscpy calls instead of strlcpy.

Link: https://lore.kernel.org/r/20210528113018.52290-8-jinpu.wang@ionos.com
Signed-off-by: Dima Stepanov <dmitrii.stepanov@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Define MIN_CHUNK_SIZE

Define MIN_CHUNK_SIZE to replace the hard-coding number.
We need 4k for metadata, so MIN_CHUNK_SIZE should be at least 8k.

Link: https://lore.kernel.org/r/20210528113018.52290-7-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Change MAX_SESS_QUEUE_DEPTH

Max IB immediate data size is 2^28 (MAX_IMM_PAYL_BITS)
and the minimum chunk size is 4096 (2^12).
Therefore the maximum sess_queue_depth is 65536 (2^16).

Link: https://lore.kernel.org/r/20210528113018.52290-6-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Clean up the code in __rtrs_srv_change_state

No need to use double switch to check the change of state everywhere,
let's change them to "if" to reduce size.

Link: https://lore.kernel.org/r/20210528113018.52290-5-jinpu.wang@ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Add error messages for cases when failing RDMA connection

It was difficult to find out why it failed to establish RDMA
connection. This patch adds some messages to show which function
has failed why.

Link: https://lore.kernel.org/r/20210528113018.52290-4-jinpu.wang@ionos.com
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-clt: Remove MAX_SESS_QUEUE_DEPTH from rtrs_send_sess_info

Client receives queue_depth value from server. There is no need
to use MAX_SESS_QUEUE_DEPTH value.

Link: https://lore.kernel.org/r/20210528113018.52290-3-jinpu.wang@ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Kill reject_w_econnreset label

We can goto reject_w_err label after initialize err with -ECONNRESET.

Link: https://lore.kernel.org/r/20210528113018.52290-2-jinpu.wang@ionos.com
Reviewed-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/srp: Use DEVICE_ATTR_*() macros

Use DEVICE_ATTR_*() helpers instead of plain DEVICE_ATTR, which makes the
code a bit shorter and easier to read.

Link: https://lore.kernel.org/r/20210528125750.20788-1-yuehaibing@huawei.com
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>