RDMA: Validate grh_required when handling AVs

Extend the existing grh_required flag to check when AV's are handled that
a GRH is present.

Since we don't want to do query_port during the AV checks for performance
reasons move the flag into the immutable_data.

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

RDMA/hfi1: Move grh_required into update_sm_ah

grh_required is intended to be a global setting where all AV's will
require a GRH, not just the sm_lid. Move the special logic to the creation
of the SM AH.

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

RDMA: Fix storage of PortInfo CapabilityMask in the kernel

The internal flag IP_BASED_GIDS was added to a field that was being used
to hold the port Info CapabilityMask without considering the effects this
will have. Since most drivers just use the value from the HW MAD it means
IP_BASED_GIDS will also become set on any HW that sets the IBA flag
IsOtherLocalChangesNoticeSupported - which is not intended.

Fix this by keeping port_cap_flags only for the IBA CapabilityMask value
and store unrelated flags externally. Move the bit definitions for this to
ib_mad.h to make it clear what is happening.

To keep the uAPI unchanged define a new set of flags in the uapi header
that are only used by ib_uverbs_query_port_resp.port_cap_flags which match
the current flags supported in rdma-core, and the values exposed by the
current kernel.

Fixes: b4a26a27287a ("IB: Report using RoCE IP based gids in port caps")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

RDMA/ipoib: Fix return code from ipoib_cm_dev_init

The proper return code is -EOPNOTSUPP and not -ENOSYS when the function
isn't supported, also make sure to return the right error code
from ipoib_transport_dev_init() when ipoib_cm_dev_init() is supported.

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

IB/core: Simplify check for RoCE route resolve

roce_resolve_route_from_path() resolves the route based on the netdevice
of the GID attribute, therefore there is no point in checking again if
the route is resolved matches the same interface it arrived.

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

IB/mlx5: Honor cnt_set_id_valid flag instead of set_id

It is incorrect to depend on set_id value to know if counters were
allocated or not. set_id_valid field is set to true when counters
were allocated. Therefore, use set_id_valid while deciding to
free counters.

Cc: <stable@vger.kernel.org> # 4.15
Fixes: aac4492ef23a ("IB/mlx5: Update counter implementation for dual port RoCE")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/mlx5: Remove unused port number parameter

Clean up a little bit code to drop unused port_num parameter.

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

IB/uverbs: Do not use uverbs_cmd_mask in the ioctl path

Instead we are now checking the function pointers directly. Get rid of
both cases in ioctl and drop the nonsense idea that destroy can fail.

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

IB/mlx5: fix uaccess beyond "count" in debugfs read/write handlers

In general, accessing userspace memory beyond the length of the supplied
buffer in VFS read/write handlers can lead to both kernel memory corruption
(via kernel_read()/kernel_write(), which can e.g. be triggered via
sys_splice()) and privilege escalation inside userspace.

In this case, the affected files are in debugfs (and should therefore only
be accessible to root), and the read handlers check that *pos is zero
(meaning that at least sys_splice() can't trigger kernel memory
corruption). Because of the root requirement, this is not a security fix,
but rather a cleanup.

For the read handlers, fix it by using simple_read_from_buffer() instead
of custom logic. Add min() calls to the write handlers.

Fixes: 4a2da0b8c078 ("IB/mlx5: Add debug control parameters for congestion control")
Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Jann Horn <jannh@google.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/rw: Fix rdma_rw_ctx_signature_init() kernel-doc header

Fixes: 0e353e34e1e7 ("IB/core: add RW API support for signature MRs")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/cxgb4: Make c4iw_poll_cq_one() easier to analyze

Introduce the function __c4iw_poll_cq_one() such that c4iw_poll_cq_one()
becomes easier to analyze for static source code analyzers. This patch
avoids that sparse reports the following:

drivers/infiniband/hw/cxgb4/cq.c:401:36: warning: context imbalance in 'c4iw_flush_hw_cq' - unexpected unlock
drivers/infiniband/hw/cxgb4/cq.c:824:9: warning: context imbalance in 'c4iw_poll_cq_one' - different lock contexts for basic block

Compile-tested only.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Steve Wise <swise@chelsio.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/cxgb3: Make iwch_poll_cq_one() easier to analyze

Introduce the function __iwch_poll_cq_one() to make iwch_poll_cq_one()
easier to analyze for static source code analyzers. This patch avoids
that sparse reports the following:

drivers/infiniband/hw/cxgb3/iwch_cq.c:187:9: warning: context imbalance in 'iwch_poll_cq_one' - different lock contexts for basic block

Compile-tested only.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Steve Wise <swise@chelsio.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/rxe: Simplify the error handling code in rxe_create_ah()

This patch not only simplifies the error handling code in rxe_create_ah()
but also removes the dead code that was left behind by commit 47ec38666210
("RDMA: Convert drivers to use sgid_attr instead of sgid_index").

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/iser: Remove set-but-not-used variables

This patch does not change any functionality.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/ocrdma: Remove a set-but-not-used variable

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/nes: Fix a compiler warning

Avoid that the following compiler warning is reported when building with
W=1:

drivers/infiniband/hw/nes/nes_hw.c:646:51: warning: suggest braces around empty body in an 'if' statement [-Wempty-body]

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Remove set-but-not-used variables

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Remove ib_find_cached_gid() and ib_find_cached_gid_by_port()

Remove these two functions since all their callers have been removed.
See also commit ea8c2d8f6014 ("RDMA/core: Remove unused ib cache
functions").

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/ipoib: Fix use of sizeof()

Make sure to use sizeof(...) instead of sizeof ... which is more
preferred.

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

RDMA/ipoib: Prefer unsigned int to bare use of unsigned

This commit replaces all the unsigned definitions in favour of 'unsigned
int' which is preferred.

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

RDMA/ipoib: Use min_t() macro instead of min()

Use min_t() macro to avoid the casting when using min() macro, also fix
the type of "length" and "wc->byte_len" to be "unsigned int" and
"u32" which is the right type for each one of them.

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

IB/cm: Remove unused and erroneous msg sequence encoding

In cm_form_tid(), a two bit message sequence number is OR'ed into bit
31-30 of the lower TID value.

After commit f06d26537559 ("IB/cm: Randomize starting comm ID"), the
local_id is XOR'ed with a 32-bit random value. Hence, bit 31-30 in the
lower TID now has an arbitrarily value and it makes no sense to OR in
the message sequence number.

Adding to that, the evolution in use of IDR routines in cm_alloc_id()
has always had the possibility of returning a value with bit 30 set.

In addition, said bits are never checked.

Hence, remove the encoding and the corresponding enum.

Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/uverbs: Replace ib_ucq_object uverbs_file with the one in ib_uobject

Now that ib_uobject has a ib_uverbs_file we don't need this extra one in
ib_ucq_object.

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

IB/uverbs: Remove ib_uobject_file

The only purpose for this structure was to hold the ib_uobject_file
pointer, but now that is part of the standard ib_uobject the structure
no longer makes any sense, so get rid of it.

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

IB/uverbs: Tidy up remaining references to ucontext

Unnecessary clutter, to indirect through ucontext when the ufile would do.
Generally most of the code code should only be working with ufile, except
for a few places that touch the driver interface.

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

IB/uverbs: Replace file->ucontext with file in uverbs_cmd.c

The ucontext isn't needed any more, just pass the uverbs_file directly.

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

IB/uverbs: Replace ib_ucontext with ib_uverbs_file in core function calls

The correct handle to refer to the idr/etc is ib_uverbs_file, revise all
the core APIs to use this instead. The user API are left as wrappers
that automatically convert a ucontext to a ufile for now.

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

IB/uverbs: Move non driver related elements from ib_ucontext to ib_ufile

The IDR is part of the ib_ufile so all the machinery to lock it, handle
closing and disassociation rightly belongs to the ufile not the ucontext.

This changes the lifetime of that data to match the lifetime of the file
descriptor which is always strictly longer than the lifetime of the
ucontext.

We need the entire locking machinery to continue to exist after ucontext
destruction to allow us to return the destroy data after a device has been
disassociated.

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

IB/uverbs: Add a uobj_perform_destroy helper

This consolidates a bunch of repeated code patterns into a helper.

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

RDMA/uverbs: Combine MIN_SZ_OR_ZERO with UVERBS_ATTR_STRUCT

After all the rework is done it is now possible to include single flags in
the type macros. Any user of UVERBS_ATTR_STRUCT needs to zero check data
past the end of the known struct to be correct, so make this mandatory,
and get rid of MIN_SZ_OR_ZERO as a user flag.

This changes UVERBS_ATTR_TYPE to refer to a struct of exact size with not
possibility of extension, convert the few users of UVERBS_ATTR_TYPE and
MIN_SZ_OR_ZERO to use UVERBS_ATTR_STRUCT.

The one user of UVERBS_ATTR_STRUCT without MIN_SZ_OR_ZERO is just
confused. There is some padding at the end of that struct, but userspace
always provides it with the padding. The construction doesn't test if the
padding is zero, so it is pointless. Just use UVERBS_ATTR_TYPE.

Finally, rename min_sz_or_zero to zero_trailing to better reflect what it
does and hopefully avoid such mis-uses in the future.

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

RDMA/uverbs: Use UVERBS_ATTR_MIN_SIZE correctly and uniformly

This newer macro allows specifying a lower bound on the accepted size, and
has an 'unlimited' upper bound. Due to this it never checks for trailing
zeroing so it doesn't make any sense to combine it with MIN_SZ_OR_ZERO, so
drop MIN_SZ_OR_ZERO when they are used together

There were a couple of places that open coded this pattern, switch them to
use the clearer UVERBS_ATTR_MIN_SIZE for clarity.

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

RDMA/uverbs: Remove UA_FLAGS

This bit of boilerplate isn't really necessary, we can use bitfields
instead of a flags enum and the macros can then individually initialize
them through the __VA_ARGS__ like everything else.

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

RDMA/uverbs: Get rid of the & in method specifications

Hide it inside the macros. The & is confusing and interferes with using
this as a generic DSL in later patches.

Since this also touches almost every line, also run the specs through
clang-format (with 'BinPackParameters: false') to make the maintenance
easier.

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

RDMA/uverbs: Simplify UVERBS_OBJECT and _TREE family of macros

Instead of the large set of indirecting macros, define the few needed
macros to directly instantiate the struct uverbs_oject_tree_def and
associated objects list.

This is small amount of code duplication but the readability is far
better.

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

RDMA/uverbs: Simplify method definition macros

Instead of the large set of indirecting macros, define the few needed
macros to directly instantiate the struct uverbs_method_def and associated
attributes list.

This is small amount of code duplication but the readability is far
better.

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

RDMA/uverbs: Simplify UVERBS_ATTR family of macros

Instead of using a complex cascade of macros, just directly provide the
initializer list each of the declarations is trying to create.

Now that the macros are simplified this also reworks the uverbs_attr_spec
to be friendly to older compilers by eliminating any unnamed
structures/unions inside, and removing the duplication of some fields. The
structure size remains at 16 bytes which was the original motivation for
some of this oddness.

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

RDMA/uverbs: Split UVERBS_ATTR_FLOW_ACTION_ESP_HANDLE

Two methods are sharing the same attribute constant, but the attribute
definitions are not the same. This should not have been done, instead
split them into two attributes with the same number.

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

RDMA/uverbs: Store the specs_root in the struct ib_uverbs_device

The specs are required to operate the uverbs file, so they belong inside
the ib_uverbs_device, not inside the ib_device. The spec passed in the
ib_device is just a communication from the driver and should not be used
during runtime.

This also changes the lifetime of the spec memory to match the
ib_uverbs_device, however at this time the spec_root can still contain
driver pointers after disassociation, so it cannot be used if ib_dev is
NULL. This is preparation for another series.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

Merge branch 'mlx5-dump-fill-mkey' into rdma.git for-next

For dependencies, branch based on 'mellanox/mlx5-next' of
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux.git

Pull Dump and fill MKEY from Leon Romanovsky:

====================
MLX5 IB HCA offers the memory key, dump_fill_mkey to increase performance,
when used in a send or receive operations.

It is used to force local HCA operations to skip the PCI bus access, while
keeping track of the processed length in the ibv_sge handling.

In this three patch series, we expose various bits in our HW spec
file (mlx5_ifc.h), move unneeded for mlx5_core FW command and export such
memory key to user space thought our mlx5-abi header file.
====================

Botched auto-merge in mlx5_ib_alloc_ucontext() resolved by hand.

* branch 'mlx5-dump-fill-mkey':
  IB/mlx5: Expose dump and fill memory key
  net/mlx5: Add hardware definitions for dump_fill_mkey
  net/mlx5: Limit scope of dump_fill_mkey function
  net/mlx5: Rate limit errors in command interface

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

IB/mlx5: Expose dump and fill memory key

MLX5 IB HCA offers the memory key, dump_fill_mkey to boost
performance, when used in a send or receive operations.

It is used to force local HCA operations to skip the PCI bus access,
while keeping track of the processed length in the ibv_sge handling.

Meaning, instead of a PCI write access the HCA leaves the target
memory untouched, and skips filling that packet section. Similar
behavior is done upon send, the HCA skips data in memory relevant
to this key and saves PCI bus access.

This functionality saves PCI read/write operations.

Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

net/mlx5: Add hardware definitions for dump_fill_mkey

MLX5 IB HCA offers the memory key, dump_fill_mkey to boost
performance by forcing local HCA operations to skip the PCI bus
access,

This patch adds needed hardware definitions.

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

net/mlx5: Limit scope of dump_fill_mkey function

mlx5_core_dump_fill_mkey() is going to be used in next
patch in IB and doesn't need to be visible to whole
mlx5_core. Move that command to mlx5_ib.

Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

RDMA/bnxt_re: Fix a bunch of off by one bugs in qplib_fp.c

The srq->swq[] is allocated in bnxt_qplib_create_srq().  It has
srq->hwq.max_elements elements so these tests should be > instead of >=
or we might go beyond the end of the array.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/bnxt_re: Fix a couple off by one bugs

The sgid_tbl->tbl[] array is allocated in bnxt_qplib_alloc_sgid_tbl().
It has sgid_tbl->max elements.  So the > should be >= to prevent
accessing one element beyond the end of the array.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/core: type promotion bug in rdma_rw_init_one_mr()

"nents" is an unsigned int, so if ib_map_mr_sg() returns a negative
error code then it's type promoted to a high unsigned int which is
treated as success.

Fixes: a060b5629ab0 ("IB/core: generic RDMA READ/WRITE API")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

MAINTAINERS: Moving out...

The 2.6.18... was a hell of a ride, and by now both me and
Roi are not dealing with iser any more. Max will replace us
as maintainer, good luck to you dear!

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Acked-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/i40w: Hold read semaphore while looking after VMA

VMA lookup is supposed to be performed while mmap_sem is held.

Fixes: f26c7c83395b ("i40iw: Add 2MB page support")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx4: Test port number before querying type.

rdma_ah_find_type() can reach into ib_device->port_immutable with a
potentially out-of-bounds port number, so check that the port number is
valid first.

Fixes: 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
Signed-off-by: Tarick Bedeir <tarick@google.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

vmw_pvrdma: Release netdev when vmxnet3 module is removed

On repeated module load/unload cycles, its possible for the pvrmda driver
to encounter this crash:

...
[  297.032448] RIP: 0010:[<ffffffff839e4620>]  [<ffffffff839e4620>] netdev_walk_all_upper_dev_rcu+0x50/0xb0
[  297.034078] RSP: 0018:ffff95087780bd08  EFLAGS: 00010286
[  297.034986] RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffff95087a0c0000
[  297.036196] RDX: ffff95087a0c0000 RSI: ffffffff839e44e0 RDI: ffff950835d0c000
[  297.037421] RBP: ffff95087780bd40 R08: ffff95087a0e0ea0 R09: abddacd03f8e0ea0
[  297.038636] R10: abddacd03f8e0ea0 R11: ffffef5901e9dbc0 R12: ffff95087a0c0000
[  297.039854] R13: ffffffff839e44e0 R14: ffff95087a0c0000 R15: ffff950835d0c828
[  297.041071] FS:  0000000000000000(0000) GS:ffff95087fc00000(0000) knlGS:0000000000000000
[  297.042443] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  297.043429] CR2: ffffffffffffffe8 CR3: 000000007a652000 CR4: 00000000003607f0
[  297.044674] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  297.045893] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  297.047109] Call Trace:
[  297.047545]  [<ffffffff839e4698>] netdev_has_upper_dev_all_rcu+0x18/0x20
[  297.048691]  [<ffffffffc05d31af>] is_eth_port_of_netdev+0x2f/0xa0 [ib_core]
[  297.049886]  [<ffffffffc05d3180>] ? is_eth_active_slave_of_bonding_rcu+0x70/0x70 [ib_core]
...

This occurs because vmw_pvrdma on probe stores a pointer to the netdev
that exists on function 0 of the same bus/device/slot (which represents
the vmxnet3 ethernet driver).  However, it never removes this pointer if
the vmxnet3 module is removed, leading to crashes resulting from use after
free dereferencing incidents like the one above.

The fix is pretty straightforward.  vmw_pvrdma should listen for
NETDEV_REGISTER and NETDEV_UNREGISTER events in its event listener code
block, and update the stored netdev pointer accordingly.  This solution
has been tested by myself and the reporter with successful results.  This
fix also allows the pvrdma driver to find its underlying ethernet device
in the event that vmxnet3 is loaded after pvrdma, which it was not able to
do before.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Reported-by: ruquin@redhat.com
Tested-by: Adit Ranadive <aditr@vmware.com>
Acked-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Fix GRE flow specification

Currently the driver sets the mask of the gre_protocol to 0xffff
without consideration in the user request.

Fix it by copy the mask from the verbs spec.

Fixes: da2f22ae7707 ("IB/mlx5: Add support for GRE flow specification")
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Ariel Levkovich <lariel@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove incorrect call to do_interrupt callback

The general interrupt handler is_rcv_avail_int() has two paths,
do_interrupt() (callback) and handle_user_interrupt().  The
do_interrupt() callback is for the threaded receive handling.
is_rcv_avail_int() cannot handle threaded IRQs.

If the do_interrupt() path is taken, and the IRQ returns
IRQ_WAKE_THREAD, the IRQ behavior will be indeterminate.

Remove incorrect call to do_interrupt() from is_rcv_avail_int(),
leaving the un-threaded (handle_user_interrupt()) path.

Fixes: f4f30031c33c ("staging/rdma/hfi1: Thread the receive interrupt.")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Set in_use_ctxts bits for user ctxts only

The in_use_ctxts bitmask is for user receive contexts only.  Setting it for
any other type of receive context is incorrect.

Move initial set of in_use_ctxts bits from the general context init to the
user context specific init. Having this bit set can allow contexts to be
incorrectly identified by some IRQ handlers. This will allow
handle_user_interrupt() will now filter user contexts correctly.

Clean up redundant is_rcv_urgent_int() user context check.

A follow on patch will clean up an incorrect code path in the
is_rcv_avail_int().

Fixes: 8737ce95c463 ("IB/hfi1: Fix an assign/ordering issue with shared context IDs")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

ib_srpt: Fix a use-after-free in __srpt_close_all_ch()

BUG: KASAN: use-after-free in srpt_set_enabled+0x1a9/0x1e0 [ib_srpt]
Read of size 4 at addr ffff8801269d23f8 by task check/29726

CPU: 4 PID: 29726 Comm: check Not tainted 4.18.0-rc2-dbg+ #4
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Call Trace:
 dump_stack+0xa4/0xf5
 print_address_description+0x6f/0x270
 kasan_report+0x241/0x360
 __asan_load4+0x78/0x80
 srpt_set_enabled+0x1a9/0x1e0 [ib_srpt]
 srpt_tpg_enable_store+0xb8/0x120 [ib_srpt]
 configfs_write_file+0x14e/0x1d0 [configfs]
 __vfs_write+0xd2/0x3b0
 vfs_write+0x101/0x270
 ksys_write+0xab/0x120
 __x64_sys_write+0x43/0x50
 do_syscall_64+0x77/0x230
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x7f235cfe6154

Fixes: aaf45bd83eba ("IB/srpt: Detect session shutdown reliably")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

ib_srpt: Fix a use-after-free in srpt_close_ch()

Avoid that KASAN reports the following:

BUG: KASAN: use-after-free in srpt_close_ch+0x4f/0x1b0 [ib_srpt]
Read of size 4 at addr ffff880151180cb8 by task check/4681

CPU: 15 PID: 4681 Comm: check Not tainted 4.18.0-rc2-dbg+ #4
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Call Trace:
 dump_stack+0xa4/0xf5
 print_address_description+0x6f/0x270
 kasan_report+0x241/0x360
 __asan_load4+0x78/0x80
 srpt_close_ch+0x4f/0x1b0 [ib_srpt]
 srpt_set_enabled+0xf7/0x1e0 [ib_srpt]
 srpt_tpg_enable_store+0xb8/0x120 [ib_srpt]
 configfs_write_file+0x14e/0x1d0 [configfs]
 __vfs_write+0xd2/0x3b0
 vfs_write+0x101/0x270
 ksys_write+0xab/0x120
 __x64_sys_write+0x43/0x50
 do_syscall_64+0x77/0x230
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Fixes: aaf45bd83eba ("IB/srpt: Detect session shutdown reliably")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/mlx5: Remove set-but-not-used variables

Avoid that the compiler complains about set-but-not-used variables when
building with W=1. This patch does not change any functionality.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/srp: Remove driver version and release data information

Remove the driver version and release date information because such
information is not relevant for an upstream driver. See also commit
e1267b01240a ("RDMA: Remove useless MODULE_VERSION").

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

include/rdma/opa_addr.h: Fix an endianness issue

IB_MULTICAST_LID_BASE is defined as follows:

  #define IB_MULTICAST_LID_BASE   cpu_to_be16(0xC000)

Hence use be16_to_cpu() to convert it to CPU endianness. Compile-tested
only.

Fixes: af808ece5ce9 ("IB/SA: Check dlid before SA agent queries for ClassPortInfo")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Venkata Sandeep Dhanalakota <venkata.s.dhanalakota@intel.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB: Improve uverbs_cleanup_ucontext algorithm

Improve uverbs_cleanup_ucontext algorithm to work properly when the
topology graph of the objects cannot be determined at compile time.  This
is the case with objects created via the devx interface in mlx5.

Typically uverbs objects must be created in a strict topologically sorted
order, so that LIFO ordering will generally cause them to be freed
properly. There are only a few cases (eg memory windows) where objects can
point to things out of the strict LIFO order.

Instead of using an explicit ordering scheme where the HW destroy is not
allowed to fail, go over the list multiple times and allow the destroy
function to fail. If progress halts then a final, desperate, cleanup is
done before leaking the memory. This indicates a driver bug.

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

IB/srpt: Support HCAs with more than two ports

Since there are adapters that have four ports, increase the size of
the srpt_device.port[] array. This patch avoids that the following
warning is hit with quad port Chelsio adapters:

    WARN_ON(sdev->device->phys_port_cnt > ARRAY_SIZE(sdev->port));

Reported-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: <stable@vger.kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/iser: set can_queue earlier to allow setting higher queue depth

We need to set can_queue earlier than when enabling the scsi host.
in a blk-mq enabled environment, the tagset allocation is taken
from can_queue which cannot be modified later. Also, pass an updated
.can_queue to iscsi_session_setup to have enough iscsi tasks allocated
in the session kfifo.

Reported-by: Karandeep Chahal <karandeepchahal@gmail.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/rxe: don't clear the tx queue on every transfer

Do not call sk_dst_set() on every packet transfer because
that calls sk_tx_queue_clear(), which clears the tx queue.
A QP must stay on the same tx queue to maintain packet order.

Signed-off-by: Vijay Immanuel <vijayi@attalasystems.com>
Acked-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/cm: Remove now useless rcu_lock in dst_fetch_ha

This lock used to be protecting a call to dst_get_neighbour_noref,
however the below commit changed it to dst_neigh_lookup which no longer
requires rcu.

Access to nud_state, neigh_event_send or rdma_copy_addr does not require
RCU, so delete the lock.

Fixes: 02b619555ad6 ("infiniband: Convert dst_fetch_ha() over to dst_neigh_lookup().")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

RDMA/mlx5: Don't leak UARs in case of free fails

The failure in releasing one UAR doesn't mean that we can't continue to
release rest of system pages, so don't return too early.

As part of cleanup, there is no need to print warning if
mlx5_cmd_free_uar() fails because such warning will be printed as part of
mlx5_cmd_exec().

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

RDMA/vmw_pvrdma: Delete unused function

This function is not in use - delete it.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Acked-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/core: Check for rdma_protocol_ib only after validating port_num

port_num is untrusted data from the user, so it should be checked after
calling fill_sgid_attr, which validates it.

Fixes: 8d9ec9addd6c ("IB/core: Add a sgid_attr pointer to struct rdma_ah_attr")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

net/mlx5: Rate limit errors in command interface

Any error status returned by FW will trigger a print similar to the
following error message in the dmesg.

[   55.884355] mlx5_core 0000:00:04.0: mlx5_cmd_check:712:(pid 555):
ALLOC_UAR(0x802) op_mod(0x0) failed, status limits exceeded(0x8),
syndrome (0x0)

Those prints are extremely valuable to diagnose issues with running system
and it is important to keep them. However, not-so-careful user can trigger
endless number of such prints by depleting HW resources and will spam
dmesg.

Rate limiting of such messages solves this issue.

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

IB/mlx4: Create slave AH's directly

Since slave GID's do not exist in the core gid table we can no longer use
the core code to help do this without creating inconsistencies. Directly
create the AH using mlx4 internal APIs.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

RDMA/uverbs: Remove redundant check

kern_spec->reserved is checked prior to calling
kern_spec_to_ib_spec_filter() which makes this second check redundant.

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

RDMA/umem: Don't check for a negative return value of dma_map_sg_attrs()

dma_map_sg_attrs() returns 0 on error and can't return a negative number
(ensured by BUG_ON), so don't check.

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

RDMA/uverbs: Don't overwrite NULL pointer with ZERO_SIZE_PTR

Number of specs is provided by user and in valid case can be equal to zero.
Such argument causes to call to kcalloc() with zero-length request and in
return the ZERO_SIZE_PTR is assigned. This pointer is different from NULL
and makes various if (..) checks to success.

Fixes: b6ba4a9aa59f ("IB/uverbs: Add support for flow counters")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/verbs: Drop kernel variant of destroy_flow

Following the removal of ib_create_flow(), adjust the code to get rid of
ib_destroy_flow() too.

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

RDMA/verbs: Drop kernel variant of create_flow

There are no kernel users of this interface so lets drop it.

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

RDMA/uverbs: Check existence of create_flow callback

In the accepted series "Refactor ib_uverbs_write path", we presented the
roadmap to get rid of uverbs_cmd_mask and uverbs_ex_cmd_mask fields in
favor of simple check of function pointer. So let's put NULL check of
create_flow function callback despite the fact that uverbs_ex_cmd_mask
still exists.

Link: https://www.spinics.net/lists/linux-rdma/msg60753.html
Suggested-by: Michael J Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

MAINTAINERS: Update SRP entries

Reflect the acquisition of SanDisk by Western Digital in my e-mail
address. Remove the reference to David Dillow's git tree since SRP patches
are queued by Doug and Jason. Remove the reference to the OpenFabrics
website since the srp_daemon source code has been moved from that website
into the rdma-core project. Add an entry for the SRP target driver.

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: David Dillow <dillow@google.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/usnic: Update with bug fixes from core code

usnic has a modified version of the core codes' ib_umem_get() and
related, and the copy misses many of the bug fixes done over the years:

Commit bc3e53f682d9 ("mm: distinguish between mlocked and pinned pages")
Commit 87773dd56d54 ("IB: ib_umem_release() should decrement mm->pinned_vm
                      from ib_umem_get")
Commit 8494057ab5e4 ("IB/uverbs: Prevent integer overflow in ib_umem_get
                      address arithmetic")
Commit 8abaae62f3fd ("IB/core: disallow registering 0-sized memory region")
Commit 66578b0b2f69 ("IB/core: don't disallow registering region starting
                      at 0x0")
Commit 53376fedb9da ("RDMA/core: not to set page dirty bit if it's already
                      set.")
Commit 8e907ed48827 ("IB/umem: Use the correct mm during ib_umem_release")

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

IB/mlx4: Add support for drain SQ & RQ

This patch follows the logic from ib_core but considers the internal
device state upon executing the involved commands.

Specifically, Upon internal error state modify QP to an error state can
be assumed to be success as each in-progress WR going to be flushed in
error in any case as expected by that modify command.

In addition,
As the drain should never fail the driver makes sure that post_send/recv
will succeed even if the device is already in an internal error state.
As such once the driver will supply the simulated/SW CQEs the CQE for
the drain WR will be handled as well.

In case of an internal error state the CQE for the drain WR may be
completed as part of the main task that handled the error state or by
the task that issued the drain WR.

As the above depends on scheduling the code takes the relevant locks
and actions to make sure that the completion handler for that WR will
always be called after that the post_send/recv were issued but not in
parallel to the other task that handles the error flow.

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

IB/mlx5: Add support for drain SQ & RQ

This patch follows the logic from ib_core but considers the internal
device state upon executing the involved commands.

Specifically,
Upon internal error state modify QP to an error state can be assumed to
be success as each in-progress WR going to be flushed in error in any
case as expected by that modify command.

In addition,
As the drain should never fail the driver makes sure that post_send/recv
will succeed even if the device is already in an internal error state.
As such once the driver will supply the simulated/SW CQEs the CQE for
the drain WR will be handled as well.

In case of an internal error state the CQE for the drain WR may be
completed as part of the main task that handled the error state or by
the task that issued the drain WR.

As the above depends on scheduling the code takes the relevant locks and
actions to make sure that the completion handler for that WR will always
be called after that the post_send/recv were issued but not in parallel
to the other task that handles the error flow.

Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/core: Remove unused ib cache functions

Now that all users have been converted to use the version of these APIs
that returns a gid_attr pointer we can delete the old entry points.

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

IB/cm: Use sgid_attr from the AV

Prior patches now ensure that the AV has a sgid_attr, if one would have
been required.  Instead of querying for one, take it directly from the AH.

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

IB/cm: Replace members of sa_path_rec with 'struct sgid_attr *'

While processing a path record entry in CM messages the associated GID
attribute is now also supplied.

Currently for RoCE a netdevice's net namespace pointer and ifindex are
stored in path record entry. Both of these fields of the netdev can change
anytime while processing CM messages. Additionally storing net namespace
without holding reference will lead to use-after-free crash. Therefore it
is removed. Netdevice information for RoCE is instead provided via
referenced gid attribute in ib_cm requests.

Such a design leads to a situation where the kernel can crash when the net
pointer becomes invalid. However today it is always initialized to
init_net, which cannot become invalid. In order to support processing
packets in any arbitrary namespace of the received packet, it is necessary
to avoid such conditions.

This patch removes the dependency on the net pointer and ifindex; instead
it will rely on SGID attribute which contains a pointer to netdev.

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

IB/cm: Pass the sgid_attr through various events

Make the sgid_attr available along with path information to the event
consumer, this allows the consumer to keep using the same GID table entry
as the event is related to.

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

IB/cm: Keep track of the sgid_attr that created the cm id

Hold reference to the the sgid_attr which is used in a cm_id until the
cm_id is destroyed.

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

IB: Make init_ah_attr_grh_fields set sgid_attr

Use the sgid and other information from the path record to figure out the
sgid_attrs.

Store the selected table entry in the sgid_attr for everything else to
use.

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

IB: Make ib_init_ah_from_mcmember set sgid_attr

This is really just a CM support function, normally a multicast address
does not have a specific SGID - but the RDMA CM usage model does restrict
things to the netdevice the CM id is bound to, at least for roce case.

Store the selected table entry in the sgid_attr for everything else to
use.

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

IB: Make ib_init_ah_attr_from_wc set sgid_attr

The work completion is inspected to determine what dgid table entry was
used to receieve the packet, produces a sgid_attr that matches and sticks
it in the ah_attr.

All callers of this function are now required to release the ah_attr on
success.

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

IB/hfi1: Remove INTx support and simplify MSIx usage

The INTx IRQ support does not work for all HF1 IRQ handlers
(specifically the receive data IRQs).

Remove all supporting code for the INTx IRQ.

If the requested MSIx vector request is unsuccessful, do not allow the
driver to continue.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Reviewed-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Reorg ctxtdata and rightsize fields

Many fields in ctxtdata are incorrectly sized and the organization of the
fields within the structure is a jumble.

Fix by:
- Correcting oversize fields.
- Putting fields common to all contexts at the top with hot fields
  at the top.
- Moving PSM fields to the bottom of the structure.

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: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove caches of chip CSRs

Remove the sizeable cache of the chip sizing CSRs and replace with CSR
reads as needed.

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: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove unused/writeonly devdata fields

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: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Rightsize ctxt_eager_bufs fields

Fields in this structure are sized excessively based on hardware
limitations and input values.

Fix by reducing fields as appropriate and repositioning to close holes in
the structure.

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: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove rcvctrl from ctxtdata

It is only ever written.

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: Jason Gunthorpe <jgg@mellanox.com>

IB/hfi1: Remove rcvhdrq_size

The usage of this ctxt data field is not hot path and the value can be
computed on demand to cut down the ctxtdata bloat.

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: Jason Gunthorpe <jgg@mellanox.com>

IB/core: Free GID table entry during GID deletion

If we already hold the table->lock when doing the kref_put it means we are
in a context where it is safe to do the deletion synchronously, with no
need for the work queue.

This helps to eliminate issues when GID change is requested as part of MAC
address change or bonding event change where expectation is to replace the
GID almost immediately.

Fixes: b150c3862d21 ("IB/core: Introduce GID entry reference counts")
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

RDMA/cma: Consider net namespace while leaving multicast group

When sending multicast leave request, consider the net ns in which this
cm_id is created.

Code was duplicated in cma_leave_mc_groups() and rdma_leave_multicast(),
which is now done using a helper function cma_leave_roce_mc_group().

Fixes: bee3c3c91865 ("IB/cma: Join and leave multicast groups with IGMP")
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

IB/uverbs: Delete type and id from uverbs_obj_attr

In this context the uobject is not allowed to be NULL, so type is the same
as uobject->type, and at least for IDR, id is the same as uobject->id.

FD objects should never handle the FD number outside the uAPI boundary
code.

Suggested-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

Merge branch 'icrc-counter' into rdma.git for-next

For dependencies, branch based on 'mellanox/mlx5-next' of
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux.git

Pull RoCE ICRC counters from Leon Romanovsky:

====================
This series exposes RoCE ICRC counter through existing RDMA hw_counters
sysfs interface.

The first patch has all HW definitions in mlx5_ifc.h file and second patch
is the actual counter implementation.
====================

* branch 'icrc-counter':
  IB/mlx5: Support RoCE ICRC encapsulated error counter
  net/mlx5: Add RoCE RX ICRC encapsulated counter

IB/mlx5: Support RoCE ICRC encapsulated error counter

This patch adds support to query the counter that counts the
RoCE packets with corrupted ICRC (Invariant Cyclic Redundancy Code).

This counter will be under
/sys/class/infiniband/<mlx5-dev>/ports/<port>/hw_counters/

rx_icrc_encapsulated - The number of RoCE packets with ICRC
error.

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

net/mlx5: Add RoCE RX ICRC encapsulated counter

Add capability bit in PCAM register and RoCE ICRC error counter
to PPCNT register.

Signed-off-by: Talat Batheesh <talatb@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

RDMA/mlx5: Refactor transport domain checks

Put all relevant checks for transport domain in the
mlx5_ib_alloc/dealloc_transport_domain functions.

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

IB/rdmavt, IB/hfi1: Create device dependent s_flags

Move some s_flags defines out of rdmavt and into hfi1 because they are
hfi1 specific and therefore should remain in the driver instead of
bubbling up to rdmavt.

Document device specific ranges in rdmavt and remap
those in hfi1.

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