RDMA/rtrs-srv: Do not pass a valid pointer to PTR_ERR()

smatch gives the warning:

  drivers/infiniband/ulp/rtrs/rtrs-srv.c:1805 rtrs_rdma_connect() warn: passing zero to 'PTR_ERR'

Which is trying to say smatch has shown that srv is not an error pointer
and thus cannot be passed to PTR_ERR.

The solution is to move the list_add() down after full initilization of
rtrs_srv. To avoid holding the srv_mutex too long, only hold it during the
list operation as suggested by Leon.

Fixes: 03e9b33a0fd6 ("RDMA/rtrs: Only allow addition of path to an already established session")
Link: https://lore.kernel.org/r/20210216143807.65923-1-jinpu.wang@cloud.ionos.com
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/srp: Fix support for unpopulated and unbalanced NUMA nodes

The current code computes a number of channels per SRP target and spreads
them equally across all online NUMA nodes.  Each channel is then assigned
a CPU within this node.

In the case of unbalanced, or even unpopulated nodes, some channels do not
get a CPU associated and thus do not get connected.  This causes the SRP
connection to fail.

This patch solves the issue by rewriting channel computation and
allocation:

- Drop channel to node/CPU association as it had no real effect on
  locality but added unnecessary complexity.

- Tweak the number of channels allocated to reduce CPU contention when
  possible:
  - Up to one channel per CPU (instead of up to 4 by node)
  - At least 4 channels per node, unless ch_count module parameter is
    used.

Link: https://lore.kernel.org/r/9cb4d9d3-30ad-2276-7eff-e85f7ddfb411@suse.com
Signed-off-by: Nicolas Morey-Chaisemartin <nmoreychaisemartin@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Merge branch 'mlx5_timestamp' into rdma.git for-next

Leon Romanovsky says:

====================
Add an extra timestamp format for mlx5_ib device.
====================

Based on the mlx5-next branch at
     git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux
due to dependencies.

Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
* branch 'mlx5_timestamp':
  RDMA/mlx5: Fail QP creation if the device can not support the CQE TS
  net/mlx5: Add new timestamp mode bits

RDMA/mlx5: Fail QP creation if the device can not support the CQE TS

In ConnectX6Dx device, HW can work in real time timestamp mode according
to the device capabilities per RQ/SQ/QP.

When the flag IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION is set, the user
expect to get TS on the CQEs in free running format, so we need to fail
the QP creation if the current mode of the device doesn't support it.

Link: https://lore.kernel.org/r/20210209131107.698833-3-leon@kernel.org
Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Signed-off-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Allow CQ creation without attached EQs

The traditional DevX CQ creation flow goes through mlx5_core_create_cq()
which checks that the given EQN corresponds to an existing EQ and attaches
a devx handler to the EQN for the CQ.

In some cases the EQ will not be a kernel EQ, but will be controlled by
modify CQ, don't block creating these just because the EQN can't be found
in the kernel.

Link: https://lore.kernel.org/r/20210211085549.1277674-1-leon@kernel.org
Signed-off-by: Tal Gilboa <talgi@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv-sysfs: fix missing put_device

put_device() decreases the ref-count and then the device will be
cleaned-up, while at is also add missing put_device in
rtrs_srv_create_once_sysfs_root_folders

This patch solves a kmemleak error as below:

  unreferenced object 0xffff88809a7a0710 (size 8):
    comm "kworker/4:1H", pid 113, jiffies 4295833049 (age 6212.380s)
    hex dump (first 8 bytes):
      62 6c 61 00 6b 6b 6b a5                          bla.kkk.
    backtrace:
      [<0000000054413611>] kstrdup+0x2e/0x60
      [<0000000078e3120a>] kobject_set_name_vargs+0x2f/0xb0
      [<00000000f1a17a6b>] dev_set_name+0xab/0xe0
      [<00000000d5502e32>] rtrs_srv_create_sess_files+0x2fb/0x314 [rtrs_server]
      [<00000000ed11a1ef>] rtrs_srv_info_req_done+0x631/0x800 [rtrs_server]
      [<000000008fc5aa8f>] __ib_process_cq+0x94/0x100 [ib_core]
      [<00000000a9599cb4>] ib_cq_poll_work+0x32/0xc0 [ib_core]
      [<00000000cfc376be>] process_one_work+0x4bc/0x980
      [<0000000016e5c96a>] worker_thread+0x78/0x5c0
      [<00000000c20b8be0>] kthread+0x191/0x1e0
      [<000000006c9c0003>] ret_from_fork+0x3a/0x50

Fixes: baa5b28b7a47 ("RDMA/rtrs-srv: Replace device_register with device_initialize and device_add")
Link: https://lore.kernel.org/r/20210212134525.103456-5-jinpu.wang@cloud.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: fix memory leak by missing kobject free

kmemleak reported an error as below:

  unreferenced object 0xffff8880674b7640 (size 64):
    comm "kworker/4:1H", pid 113, jiffies 4296403507 (age 507.840s)
    hex dump (first 32 bytes):
      69 70 3a 31 39 32 2e 31 36 38 2e 31 32 32 2e 31  ip:192.168.122.1
      31 30 40 69 70 3a 31 39 32 2e 31 36 38 2e 31 32  10@ip:192.168.12
    backtrace:
      [<0000000054413611>] kstrdup+0x2e/0x60
      [<0000000078e3120a>] kobject_set_name_vargs+0x2f/0xb0
      [<00000000ca2be3ee>] kobject_init_and_add+0xb0/0x120
      [<0000000062ba5e78>] rtrs_srv_create_sess_files+0x14c/0x314 [rtrs_server]
      [<00000000b45b7217>] rtrs_srv_info_req_done+0x5b1/0x800 [rtrs_server]
      [<000000008fc5aa8f>] __ib_process_cq+0x94/0x100 [ib_core]
      [<00000000a9599cb4>] ib_cq_poll_work+0x32/0xc0 [ib_core]
      [<00000000cfc376be>] process_one_work+0x4bc/0x980
      [<0000000016e5c96a>] worker_thread+0x78/0x5c0
      [<00000000c20b8be0>] kthread+0x191/0x1e0
      [<000000006c9c0003>] ret_from_fork+0x3a/0x50

It is caused by the not-freed kobject of rtrs_srv_sess.  The kobject
embedded in rtrs_srv_sess has ref-counter 2 after calling
process_info_req(). Therefore it must call kobject_put twice.  Currently
it calls kobject_put only once at rtrs_srv_destroy_sess_files because
kobject_del removes the state_in_sysfs flag and then kobject_put in
free_sess() is not called.

This patch moves kobject_del() into free_sess() so that the kobject of
rtrs_srv_sess can be freed. And also this patch adds the missing call of
sysfs_remove_group() to clean-up the sysfs directory.

Fixes: 9cb837480424 ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20210212134525.103456-4-jinpu.wang@cloud.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Only allow addition of path to an already established session

While adding a path from the client side to an already established
session, it was possible to provide the destination IP to a different
server. This is dangerous.

This commit adds an extra member to the rtrs_msg_conn_req structure, named
first_conn; which is supposed to notify if the connection request is the
first for that session or not.

On the server side, if a session does not exist but the first_conn
received inside the rtrs_msg_conn_req structure is 1, the connection
request is failed. This signifies that the connection request is for an
already existing session, and since the server did not find one, it is an
wrong connection request.

Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
Fixes: 9cb837480424 ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20210212134525.103456-3-jinpu.wang@cloud.ionos.com
Signed-off-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Reviewed-by: Lutz Pogrell <lutz.pogrell@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs-srv: Fix stack-out-of-bounds

  BUG: KASAN: stack-out-of-bounds in _mlx4_ib_post_send+0x1bd2/0x2770 [mlx4_ib]
  Read of size 4 at addr ffff8880d5a7f980 by task kworker/0:1H/565

  CPU: 0 PID: 565 Comm: kworker/0:1H Tainted: G           O      5.4.84-storage #5.4.84-1+feature+linux+5.4.y+dbg+20201216.1319+b6b887b~deb10
  Hardware name: Supermicro H8QG6/H8QG6, BIOS 3.00       09/04/2012
  Workqueue: ib-comp-wq ib_cq_poll_work [ib_core]
  Call Trace:
   dump_stack+0x96/0xe0
   print_address_description.constprop.4+0x1f/0x300
   ? irq_work_claim+0x2e/0x50
   __kasan_report.cold.8+0x78/0x92
   ? _mlx4_ib_post_send+0x1bd2/0x2770 [mlx4_ib]
   kasan_report+0x10/0x20
   _mlx4_ib_post_send+0x1bd2/0x2770 [mlx4_ib]
   ? check_chain_key+0x1d7/0x2e0
   ? _mlx4_ib_post_recv+0x630/0x630 [mlx4_ib]
   ? lockdep_hardirqs_on+0x1a8/0x290
   ? stack_depot_save+0x218/0x56e
   ? do_profile_hits.isra.6.cold.13+0x1d/0x1d
   ? check_chain_key+0x1d7/0x2e0
   ? save_stack+0x4d/0x80
   ? save_stack+0x19/0x80
   ? __kasan_slab_free+0x125/0x170
   ? kfree+0xe7/0x3b0
   rdma_write_sg+0x5b0/0x950 [rtrs_server]

The problem is when we send imm_wr, the type should be ib_rdma_wr, so hw
driver like mlx4 can do rdma_wr(wr), so fix it by use the ib_rdma_wr as
type for imm_wr.

Fixes: 9cb837480424 ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20210212134525.103456-2-jinpu.wang@cloud.ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Remove unused pkt->offset

The pkt->offset field is never used except to assign it to 0.  But it adds
lots of unneeded code. This patch removes the field and related code. This
causes a measurable improvement in performance.

Link: https://lore.kernel.org/r/20210211210455.3274-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/ucma: Fix use-after-free bug in ucma_create_uevent

ucma_process_join() allocates struct ucma_multicast mc and frees it if an
error occurs during its run.  Specifically, if an error occurs in
copy_to_user(), a use-after-free might happen in the following scenario:

1. mc struct is allocated.
2. rdma_join_multicast() is called and succeeds. During its run,
   cma_iboe_join_multicast() enqueues a work that will later use the
   aforementioned mc struct.
3. copy_to_user() is called and fails.
4. mc struct is deallocated.
5. The work that was enqueued by cma_iboe_join_multicast() is run and
   calls ucma_create_uevent() which tries to access mc struct (which is
   freed by now).

Fix this bug by cancelling the work enqueued by cma_iboe_join_multicast().
Since cma_work_handler() frees struct cma_work, we don't use it in
cma_iboe_join_multicast() so we can safely cancel the work later.

The following syzkaller report revealed it:

   BUG: KASAN: use-after-free in ucma_create_uevent+0x2dd/0x;3f0 drivers/infiniband/core/ucma.c:272
   Read of size 8 at addr ffff88810b3ad110 by task kworker/u8:1/108

   CPU: 1 PID: 108 Comm: kworker/u8:1 Not tainted 5.10.0-rc6+ #257
   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS   rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
   Workqueue: rdma_cm cma_work_handler
   Call Trace:
    __dump_stack lib/dump_stack.c:77 [inline]
    dump_stack+0xbe/0xf9 lib/dump_stack.c:118
    print_address_description.constprop.0+0x3e/0×60 mm/kasan/report.c:385
    __kasan_report mm/kasan/report.c:545 [inline]
    kasan_report.cold+0x1f/0×37 mm/kasan/report.c:562
    ucma_create_uevent+0x2dd/0×3f0 drivers/infiniband/core/ucma.c:272
    ucma_event_handler+0xb7/0×3c0 drivers/infiniband/core/ucma.c:349
    cma_cm_event_handler+0x5d/0×1c0 drivers/infiniband/core/cma.c:1977
    cma_work_handler+0xfa/0×190 drivers/infiniband/core/cma.c:2718
    process_one_work+0x54c/0×930 kernel/workqueue.c:2272
    worker_thread+0x82/0×830 kernel/workqueue.c:2418
    kthread+0x1ca/0×220 kernel/kthread.c:292
    ret_from_fork+0x1f/0×30 arch/x86/entry/entry_64.S:296

   Allocated by task 359:
     kasan_save_stack+0x1b/0×40 mm/kasan/common.c:48
     kasan_set_track mm/kasan/common.c:56 [inline]
     __kasan_kmalloc mm/kasan/common.c:461 [inline]
     __kasan_kmalloc.constprop.0+0xc2/0xd0 mm/kasan/common.c:434
     kmalloc include/linux/slab.h:552 [inline]
     kzalloc include/linux/slab.h:664 [inline]
     ucma_process_join+0x16e/0×3f0 drivers/infiniband/core/ucma.c:1453
     ucma_join_multicast+0xda/0×140 drivers/infiniband/core/ucma.c:1538
     ucma_write+0x1f7/0×280 drivers/infiniband/core/ucma.c:1724
     vfs_write fs/read_write.c:603 [inline]
     vfs_write+0x191/0×4c0 fs/read_write.c:585
     ksys_write+0x1a1/0×1e0 fs/read_write.c:658
     do_syscall_64+0x2d/0×40 arch/x86/entry/common.c:46
     entry_SYSCALL_64_after_hwframe+0x44/0xa9

   Freed by task 359:
     kasan_save_stack+0x1b/0×40 mm/kasan/common.c:48
     kasan_set_track+0x1c/0×30 mm/kasan/common.c:56
     kasan_set_free_info+0x1b/0×30 mm/kasan/generic.c:355
     __kasan_slab_free+0x112/0×160 mm/kasan/common.c:422
     slab_free_hook mm/slub.c:1544 [inline]
     slab_free_freelist_hook mm/slub.c:1577 [inline]
     slab_free mm/slub.c:3142 [inline]
     kfree+0xb3/0×3e0 mm/slub.c:4124
     ucma_process_join+0x22d/0×3f0 drivers/infiniband/core/ucma.c:1497
     ucma_join_multicast+0xda/0×140 drivers/infiniband/core/ucma.c:1538
     ucma_write+0x1f7/0×280 drivers/infiniband/core/ucma.c:1724
     vfs_write fs/read_write.c:603 [inline]
     vfs_write+0x191/0×4c0 fs/read_write.c:585
     ksys_write+0x1a1/0×1e0 fs/read_write.c:658
     do_syscall_64+0x2d/0×40 arch/x86/entry/common.c:46
     entry_SYSCALL_64_after_hwframe+0x44/0xa9
     The buggy address belongs to the object at ffff88810b3ad100
     which belongs to the cache kmalloc-192 of size 192
     The buggy address is located 16 bytes inside of
     192-byte region [ffff88810b3ad100, ffff88810b3ad1c0)

Fixes: b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join")
Link: https://lore.kernel.org/r/20210211090517.1278415-1-leon@kernel.org
Reported-by: Amit Matityahu <mitm@nvidia.com>
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Fix kernel doc warnings for ib_port_immutable_read()

drivers/infiniband/core/device.c:859: warning: Function parameter or member 'dev' not described in 'ib_port_immutable_read'
drivers/infiniband/core/device.c:859: warning: Function parameter or member 'port' not described in 'ib_port_immutable_read'

Fixes: 7416790e2245 ("RDMA/core: Introduce and use API to read port immutable data")
Link: https://lore.kernel.org/r/20210210151421.1108809-1-leon@kernel.org
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/qedr: Use true and false for bool variable

Fix the following coccicheck warning:

./drivers/infiniband/hw/qedr/qedr.h:629:9-10: WARNING: return of 0/1 in
function 'qedr_qp_has_rq' with return type bool.

./drivers/infiniband/hw/qedr/qedr.h:620:9-10: WARNING: return of 0/1 in
function 'qedr_qp_has_sq' with return type bool.

Link: https://lore.kernel.org/r/1612949901-109873-1-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>
Acked-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Adjust definition of FRMR fields

FRMR is not well-supported on HIP08, it is re-designed for HIP09 and the
position of related fields is changed. Then the ULPs should be forbidden
to use FRMR on older hardwares.

Link: https://lore.kernel.org/r/1612924424-28217-1-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Refactor process of posting CMDQ

Simplify __hns_roce_cmq_send() then remove the redundant variables.

Link: https://lore.kernel.org/r/1612688143-28226-6-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Adjust fields and variables about CMDQ tail/head

The register 0x07014 is actually the head pointer of CMDQ, and 0x07010
means tail pointer. Current definitions are confusing, so rename them and
related variables.

The next_to_use of structure hns_roce_v2_cmq_ring has the same semantics
as head, merge them into one member. The next_to_clean of structure
hns_roce_v2_cmq_ring has the same semantics as tail. After deleting
next_to_clean, tail should also be deleted.

Link: https://lore.kernel.org/r/1612688143-28226-5-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove redundant operations on CMDQ

CMDQ works serially, after each successful transmission, the head and tail
pointers will be equal, so there is no need to check whether the queue is
full. At the same time, since the descriptor of each transmission is new,
there is no need to perform a cleanup operation. Then, the field named
next_to_clean in structure hns_roce_v2_cmq_ring is redundant.

Fixes: a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
Link: https://lore.kernel.org/r/1612688143-28226-4-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Fixes missing error code of CMDQ

When posting a multi-descriptors command, the error code of previous
failed descriptors may be rewrote to 0 by a later successful descriptor.

Fixes: a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
Link: https://lore.kernel.org/r/1612688143-28226-3-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove unused member and variable of CMDQ

last_status of structure hns_roce_v2_cmq has never been used, and the
variable named 'complete' in __hns_roce_cmq_send() is meaningless.

Fixes: a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
Link: https://lore.kernel.org/r/1612688143-28226-2-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/ipoib: Remove racy Subnet Manager sendonly join checks

When a system receives a REREG event from the SM, then the SM information
in the kernel is marked as invalid and a request is sent to the SM to
update the information. The SM information is invalid in that time period.

However, receiving a REREG also occurs simultaneously in user space
applications that are now trying to rejoin the multicast groups. Some of
those may be sendonly multicast groups which are then failing.

If the SM information is invalid then ib_sa_sendonly_fullmem_support()
returns false. That is wrong because it just means that we do not know yet
if the potentially new SM supports sendonly joins.

Sendonly join was introduced in 2015 and all the Subnet managers have
supported it ever since. So there is no point in checking if a subnet
manager supports it.

Should an old opensm get a request for a sendonly join then the request
will fail. The code that is removed here accomodated that situation and
fell back to a full join.

Falling back to a full join is problematic in itself. The reason to use
the sendonly join was to reduce the traffic on the Infiniband fabric
otherwise one could have just stayed with the regular join.  So this patch
may cause users of very old opensms to discover that lots of traffic
needlessly crosses their IB fabrics.

Link: https://lore.kernel.org/r/alpine.DEB.2.22.394.2101281845160.13303@www.lameter.com
Signed-off-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

net/mlx5: Add new timestamp mode bits

These fields declare which timestamp mode is supported by the device
per RQ/SQ/QP.

In addition add the ts_format field to the select the mode for
RQ/SQ/QP.

Link: https://lore.kernel.org/r/20210209131107.698833-2-leon@kernel.org
Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Signed-off-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

RDMA/mlx5: Support 400Gbps IB rate in mlx5 driver

Support 400Gbps IB rate in mlx5 driver.

Link: https://lore.kernel.org/r/20210209130429.698237-1-leon@kernel.org
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/qedr: Remove in_irq() usage from debug output

qedr_gsi_post_send() has a debug output which prints the return value of
in_irq() and irqs_disabled().

The result of the in_irq(), even if invoked from an interrupt handler, is
subject to change depending on the `threadirqs' command line switch.  The
result of irqs_disabled() is always be 1 because the function acquires
spinlock_t with spin_lock_irqsave().

Remove in_irq() and irqs_disabled() from the debug output because it
provides little value.

Link: https://lore.kernel.org/r/20210208193347.383254-1-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Cleanup init_send_wqe

This patch changes the type of init_send_wqe in rxe_verbs.c to void since
it always returns 0. It also separates out the code that copies inline
data into the send wqe as copy_inline_data_to_wqe().

Link: https://lore.kernel.org/r/20210206002437.2756-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix minor coding style issues

checkpatch -f found 3 warnings in RDMA/rxe

1. a missing space following switch
2. return followed by else
3. use of strlcpy() instead of strscpy().

This patch fixes each of these. In

...
} elseif (...) {
...
return 0;
} else
...

The middle block can be safely moved since it is completely independent of
the other code.

Link: https://lore.kernel.org/r/20210205230525.49068-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Cleanup the synchronize_srcu() from the ODP flow

Cleanup the synchronize_srcu() from the ODP flow as it was found to be a
very heavy time consumer as part of dereg_mr.

For example de-registration of 10000 ODP MRs each with size of 2M hugepage
took 19.6 sec comparing de-registration of same number of non ODP MRs that
took 172 ms.

The new locking scheme uses the wait_event() mechanism which follows the
use count of the MR instead of using synchronize_srcu().

By that change, the time required for the above test took 95 ms which is
even better than the non ODP flow.

Once fully dropped the srcu usage, had to come with a lock to protect the
XA access.

As part of using the above mechanism we could also clean the
num_deferred_work stuff and follow the use count instead.

Link: https://lore.kernel.org/r/20210202071309.2057998-1-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Delete redundant judgment when preparing descriptors

There is no need to use a for loop to assign values for an array of cmd
descriptors which has only two elements.

Link: https://lore.kernel.org/r/1612517974-31867-13-git-send-email-liweihang@huawei.com
Signed-off-by: Xinhao Liu <liuxinhao5@hisilicon.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove unnecessary wrap around for EQ's consumer index

The hns driver wrap around the consumer index of AEQ and CEQ when they
reach to two times of queue entries number for owner mechanism, actually,
it is unnecessary to wrap around since the hardware itself will mask it
before use.

Link: https://lore.kernel.org/r/1612517974-31867-12-git-send-email-liweihang@huawei.com
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Avoid unnecessary memset on WQEs in post_send

All fields of WQE will be rewrote, so the memset is unnecessary. And when
SQ is working in OWNER mode, the pipeline may prefetch the WQEs beyond PI,
the memset operation may flip the owner bit too early, then the pipeline
may get a wrong WQ.

Link: https://lore.kernel.org/r/1612517974-31867-11-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove some magic numbers

Use macros instead of magic numbers to represent shift of dma_handle_wqe,
dma_handle_idx and UDP destination port number of RoCEv2.

Link: https://lore.kernel.org/r/1612517974-31867-10-git-send-email-liweihang@huawei.com
Signed-off-by: Xinhao Liu <liuxinhao5@hisilicon.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Move HIP06 related definitions into hns_roce_hw_v1.h

hns_roce_device.h is not specific to hardware, some definitions are only
used for HIP06, they should be moved into hns_roce_hw_v1.h.

Link: https://lore.kernel.org/r/1612517974-31867-9-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Replace wmb&__raw_writeq with writeq

Currently, the driver updates doorbell looks like this:

post()
{
wqe.field = 0x111;
wmb();
update_wq_db();
}

update_wq_db()
{
db.field = 0x222;
__raw_writeq(db, db_reg);
}

writeq() is a better choice than __raw_writeq() because it calls dma_wmb()
to barrier in ARM64, and dma_wmb() is better than wmb() for ROCEE device.

This patch removes all wmb() before updating doorbell of SQ/RQ/CQ/SRQ by
replacing __raw_writeq() with writeq() to improve performence.  The new
process looks like this:

post()
{
wqe.field = 0x111;
update_wq_db();
}

update_wq_db()
{
db.field = 0x222;
writeq(db, db_reg);
}

Link: https://lore.kernel.org/r/1612517974-31867-8-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Skip qp_flow_control_init() for HIP09

Since HIP09 does not require this function, it should be masked.

Link: https://lore.kernel.org/r/1612517974-31867-7-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Disable RQ inline by default

This feature should only be enabled by querying capability from firmware.

Fixes: ba6bb7e97421 ("RDMA/hns: Add interfaces to get pf capabilities from firmware")
Link: https://lore.kernel.org/r/1612517974-31867-5-git-send-email-liweihang@huawei.com
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add mapped page count checking for MTR

Add the mapped page count checking flow to avoid invalid page size when
creating MTR.

Fixes: 38389eaa4db1 ("RDMA/hns: Add mtr support for mixed multihop addressing")
Link: https://lore.kernel.org/r/1612517974-31867-4-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/hns: Fix type of sq_signal_bits

This bit should be in type of enum ib_sig_type, or there will be a sparse
warning.

Fixes: bfe860351e31 ("RDMA/hns: Fix cast from or to restricted __le32 for driver")
Link: https://lore.kernel.org/r/1612517974-31867-3-git-send-email-liweihang@huawei.com
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Avoid filling sgid index when modifying QP to RTR

ULP usually set IB(V)_QP_AV when trying to modify QP to RTR if they want
to record sgid index into QPC. For UD QPs, it is useless because it will
be included in WQE. For RC QPs, it will be filled in
hns_roce_set_path(). So sgid index shouldn't be filled by default. Then
hns_get_gid_index() is moved to hns_roce_hw_v1.c because it is only called
in it.

Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/1612517974-31867-2-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add support of direct wqe

Direct wqe is a mechanism to fill wqe directly into the hardware. In the
case of light load, the wqe will be filled into pcie bar space of the
hardware, this will reduce one memory access operation and therefore
reduce the latency.

Link: https://lore.kernel.org/r/1611997513-27107-1-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/siw: Fix calculation of tx_valid_cpus size

The size of tx_valid_cpus was calculated under the assumption that the
numa nodes identifiers are continuous, which is not the case in all archs
as this could lead to the following panic when trying to access an invalid
tx_valid_cpus index, avoid the following panic by using nr_node_ids
instead of num_online_nodes() to allocate the tx_valid_cpus size.

   Kernel attempted to read user page (8) - exploit attempt? (uid: 0)
   BUG: Kernel NULL pointer dereference on read at 0x00000008
   Faulting instruction address: 0xc0080000081b4a90
   Oops: Kernel access of bad area, sig: 11 [#1]
   LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV
   Modules linked in: siw(+) rfkill rpcrdma ib_isert iscsi_target_mod ib_iser libiscsi scsi_transport_iscsi ib_srpt target_core_mod ib_srp scsi_transport_srp ib_ipoib rdma_ucm sunrpc ib_umad rdma_cm ib_cm iw_cm i40iw ib_uverbs ib_core i40e ses enclosure scsi_transport_sas ipmi_powernv ibmpowernv at24 ofpart ipmi_devintf regmap_i2c ipmi_msghandler powernv_flash uio_pdrv_genirq uio mtd opal_prd zram ip_tables xfs libcrc32c sd_mod t10_pi ast i2c_algo_bit drm_vram_helper drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops cec drm_ttm_helper ttm drm vmx_crypto aacraid drm_panel_orientation_quirks dm_mod
   CPU: 40 PID: 3279 Comm: modprobe Tainted: G        W      X --------- ---  5.11.0-0.rc4.129.eln108.ppc64le #2
   NIP:  c0080000081b4a90 LR: c0080000081b4a2c CTR: c0000000007ce1c0
   REGS: c000000027fa77b0 TRAP: 0300   Tainted: G        W      X --------- ---   (5.11.0-0.rc4.129.eln108.ppc64le)
   MSR:  9000000002009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE>  CR: 44224882  XER: 00000000
   CFAR: c0000000007ce200 DAR: 0000000000000008 DSISR: 40000000 IRQMASK: 0
   GPR00: c0080000081b4a2c c000000027fa7a50 c0080000081c3900 0000000000000040
   GPR04: c000000002023080 c000000012e1c300 000020072ad70000 0000000000000001
   GPR08: c000000001726068 0000000000000008 0000000000000008 c0080000081b5758
   GPR12: c0000000007ce1c0 c0000007fffc3000 00000001590b1e40 0000000000000000
   GPR16: 0000000000000000 0000000000000001 000000011ad68fc8 00007fffcc09c5c8
   GPR20: 0000000000000008 0000000000000000 00000001590b2850 00000001590b1d30
   GPR24: 0000000000043d68 000000011ad67a80 000000011ad67a80 0000000000100000
   GPR28: c000000012e1c300 c0000000020271c8 0000000000000001 c0080000081bf608
   NIP [c0080000081b4a90] siw_init_cpulist+0x194/0x214 [siw]
   LR [c0080000081b4a2c] siw_init_cpulist+0x130/0x214 [siw]
   Call Trace:
   [c000000027fa7a50] [c0080000081b4a2c] siw_init_cpulist+0x130/0x214 [siw] (unreliable)
   [c000000027fa7a90] [c0080000081b4e68] siw_init_module+0x40/0x2a0 [siw]
   [c000000027fa7b30] [c0000000000124f4] do_one_initcall+0x84/0x2e0
   [c000000027fa7c00] [c000000000267ffc] do_init_module+0x7c/0x350
   [c000000027fa7c90] [c00000000026a180] __do_sys_init_module+0x210/0x250
   [c000000027fa7db0] [c0000000000387e4] system_call_exception+0x134/0x230
   [c000000027fa7e10] [c00000000000d660] system_call_common+0xf0/0x27c
   Instruction dump:
   40810044 3d420000 e8bf0000 e88a82d0 3d420000 e90a82c8 792a1f24 7cc4302a
   7d2642aa 79291f24 7d25482a 7d295214 <7d4048a8> 7d4a3b78 7d4049ad 40c2fff4

Fixes: bdcf26bf9b3a ("rdma/siw: network and RDMA core interface")
Link: https://lore.kernel.org/r/20210201112922.141085-1-kamalheib1@gmail.com
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Reviewed-by: Bernard Metzler <bmt@zurich.ibm.com>
Tested-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add verification of QP type when post_recv

The post_recv only supports QP types of RC, GSI and UD.

Link: https://lore.kernel.org/r/1611997090-48820-13-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Refactor hns_roce_v2_post_srq_recv()

The SRQ in the hns driver consists of the following four parts:

* wqe buf: the buffer to store WQE.

* wqe_idx buf: the cqe of SRQ may be not generated in the order of wqe, so
  the wqe_idx corresponding to the idle WQE needs to be pushed into the
  index queue which is a FIFO, then it instructs the hardware to obtain
  the corresponding WQE.

* bitmap: bitmap is used to generate and release wqe_idx. When the user
  has a new WR, the driver finds the idx of the idle wqe in bitmap. When
  the CQE of wqe is generated, the driver will release the idx.

* wr_id buf: wr_id buf is used to store the user's wr_id, then return it
  to the user when poll_cq verb is invoked.

The process of post SRQ recv is refactored to make preceding code clearer.

Link: https://lore.kernel.org/r/1611997090-48820-12-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Clear remaining unused sges when post_recv

The HIP09 requires the driver to clear the unused data segments in wqe
buffer to make the hns ROCEE stop reading the remaining invalid sges for
RQ.

Link: https://lore.kernel.org/r/1611997090-48820-11-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/hns: Refactor post recv flow

Refactor post recv flow by removing unnecessary checking and removing
duplicated code.

Link: https://lore.kernel.org/r/1611997090-48820-10-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/hns: Use new interfaces to write SRQC

Use new register operation interfaces to simplify the process of write SRQ
Context.

Link: https://lore.kernel.org/r/1611997090-48820-9-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Refactor code about SRQ Context

Reduce parameter numbers of write_srqc() and move some related code into
it from alloc_srqc().

Link: https://lore.kernel.org/r/1611997090-48820-8-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Refactor hns_roce_create_srq()

Split the SRQ creation process into multiple steps and encapsulate them
into functions.

Link: https://lore.kernel.org/r/1611997090-48820-7-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove the reserved WQE of SRQ

Each SRQs contain an reserved WQE, it is inappropriate and should be
removed.

Fixes: c7bcb13442e1 ("RDMA/hns: Add SRQ support for hip08 kernel mode")
Link: https://lore.kernel.org/r/1611997090-48820-6-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Fixed wrong judgments in the goto branch

When an error occurs, the qp_table must be cleared, regardless of whether
the SRQ feature is enabled.

Fixes: 5c1f167af112 ("RDMA/hns: Init SRQ table for hip08")
Link: https://lore.kernel.org/r/1611997090-48820-5-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Force srq_limit to 0 when creating SRQ

According to the IB Specification, srq_limit shouldn't be configured
during SRQ creation. If a user set srq_limit at this time, the driver
should forced it to zero, or the result of creating SRQ will conflict with
the result of querying SRQ.

Fixes: c7bcb13442e1 ("RDMA/hns: Add SRQ support for hip08 kernel mode")
Link: https://lore.kernel.org/r/1611997090-48820-4-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Bugfix for checking whether the srq is full when post wr

If a user posts WR by wr_list, the head pointer of idx_queue won't be
updated until all wqes are filled, so the judgment of whether head equals
to tail will get a wrong result. Fix above issue and move the head and
tail pointer from the srq structure into the idx_queue structure. After
idx_queue is filled with wqe idx, the head pointer of it will increase.

Fixes: c7bcb13442e1 ("RDMA/hns: Add SRQ support for hip08 kernel mode")
Link: https://lore.kernel.org/r/1611997090-48820-3-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Allocate one more recv SGE for HIP08

The RQ/SRQ of HIP08 needs one special sge to stop receive reliably. So the
driver needs to allocate at least one SGE when creating RQ/SRQ and ensure
that at least one SGE is filled with the special value during post_recv.

Besides, the kernel driver should only do this for kernel ULP. For
userspace ULP, the userspace driver will allocate the reserved SGE in
buffer, and the kernel driver just needs to pin the corresponding size of
memory based on the userspace driver's requirements.

Link: https://lore.kernel.org/r/1611997090-48820-2-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix FIXME in rxe_udp_encap_recv()

rxe_udp_encap_recv() drops the reference to rxe->ib_dev taken by
rxe_get_dev_from_net() which should be held until each received skb is
freed. This patch moves the calls to ib_device_put() to each place a
received skb is freed. It also takes references to the ib_device for each
cloned skb created to process received multicast packets.

Fixes: 4c173f596b3f ("RDMA/rxe: Use ib_device_get_by_netdev() instead of open coding")
Link: https://lore.kernel.org/r/20210128233318.2591-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Correct skb on loopback path

rxe_net.c sends packets at the IP layer with skb->data pointing at the IP
header but receives packets from a UDP tunnel with skb->data pointing at
the UDP header.  On the loopback path this was not correctly accounted
for.  This patch corrects for this by using sbk_pull() to strip the IP
header from the skb on received packets.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Link: https://lore.kernel.org/r/20210128182301.16859-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix coding error in rxe_rcv_mcast_pkt

rxe_rcv_mcast_pkt() in rxe_recv.c can leak SKBs in error path code. The
loop over the QPs attached to a multicast group creates new cloned SKBs
for all but the last QP in the list and passes the SKB and its clones to
rxe_rcv_pkt() for further processing. Any QPs that do not pass some checks
are skipped.  If the last QP in the list fails the tests the SKB is
leaked.  This patch checks if the SKB for the last QP was used and if not
frees it. Also removes a redundant loop invariant assignment.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Fixes: 71abf20b28ff ("RDMA/rxe: Handle skb_clone() failure in rxe_recv.c")
Link: https://lore.kernel.org/r/20210128174752.16128-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Remove useless code in rxe_recv.c

In check_keys() in rxe_recv.c

if ((...) && pkt->mask) {
...
}

always has pkt->mask non zero since in rxe_udp_encap_recv() pkt->mask is
always set to RXE_GRH_MASK (!= 0).  There is no obvious reason for this
additional test and the original intent is lost. This patch simplifies the
expression.

Fixes: 8b7b59d030cc ("IB/rxe: remove redudant qpn check")
Link: https://lore.kernel.org/r/20210127224203.2812-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix coding error in rxe_recv.c

check_type_state() in rxe_recv.c is written as if the type bits in the
packet opcode were a bit mask which is not correct. This patch corrects
this code to compare all 3 type bits to the required type.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Link: https://lore.kernel.org/r/20210127214500.3707-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx5: Use rdma_for_each_port for port iteration

Instead of open coding the loop for port iteration, use rdma_for_each_port
macro provided by core.

To use such macro, early initialization of phys_port_cnt is needed.
Hence, initialize such constant early enough in the init stage.

Whichever functions are called with port using rdma_for_each_port(),
convert their port type from u8 to unsigned int to match the core API.

Link: https://lore.kernel.org/r/20210203130133.4057329-6-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Introduce and use API to read port immutable data

Currently mlx5 driver caches port GID table length for 2 ports.  It is
also cached by IB core as port immutable data.

When mlx5 representor ports are present, which are usually more than 2,
invalid access to port_caps array can happen while validating the GID
table length which is only for 2 ports.

To avoid this, take help of the IB cores port immutable data by exposing
an API to read the port immutable fields.

Remove mlx5 driver's internal cache, thereby reduce code and data.

Link: https://lore.kernel.org/r/20210203130133.4057329-5-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx5: Improve query port for representor port

Improve query port functionality for representor port as below.

1. RoCE Qkey violation counters are not applicable for representor port.
2. Avoid setting gid_tbl_len twice for representor port.
3. Avoid setting ip_gids and IB_PORT_CM_SUP property for representor port
   as GID table is empty and CM support is not present in representor
   mode.

Link: https://lore.kernel.org/r/20210203130133.4057329-4-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx5: Avoid calling query device for reading pkey table length

Pkey table length for all the ports of the device is the same.  Currently
get_ports_cap() reads and stores it for each port by querying the device
which reads more than just pkey table length.

For representor device ports which can be in range of hundreds, it queries
is for each such port and end up returning same value for all the ports.

When in representor mode, modify QP accesses pkey port caps for a port
index that can be outside of the port_caps table.

Hence, simplify the logic to query the max pkey table length only once
during device initialization sequence.

Link: https://lore.kernel.org/r/20210203130133.4057329-3-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx5: Move mlx5_port_caps from mlx5_core_dev to mlx5_ib_dev

mlx5_port_caps are RDMA specific capabilities. These are not used by the
mlx5_core_device at all. Move them to mlx5_ib_dev where it is used and
reduce the scope of it to multiple drivers.

Link: https://lore.kernel.org/r/20210203130133.4057329-2-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/pvrdma: Replace spin_lock_irqsave with spin_lock in hard IRQ

There is no need to do irqsave and irqrestore in context of hard IRQ.

Link: https://lore.kernel.org/r/1612514278-49220-1-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx5: Support default partition key for representor port

Representor port has only one default pkey.  Hence have simpler query pkey
callback or it.

Link: https://lore.kernel.org/r/20210127150010.1876121-4-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/core: Use valid port number to check link layer

IB HCA port starts from 1. Use IB core provided port iterator API to avoid
any assumption with start port number.

Link: https://lore.kernel.org/r/20210127150010.1876121-11-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx4: Use port iterator and validation APIs

Use IB core provided routine to check if the port is valid or not and to
iterate over IB ports.

Link: https://lore.kernel.org/r/20210127150010.1876121-10-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/cm: Avoid a loop when device has 255 ports

When RDMA device has 255 ports, loop iterator i overflows.  Due to which
cm_add_one() port iterator loops infinitely.  Use core provided port
iterator to avoid the infinite loop.

Fixes: a977049dacde ("[PATCH] IB: Add the kernel CM implementation")
Link: https://lore.kernel.org/r/20210127150010.1876121-9-leon@kernel.org
Signed-off-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/mlx5: Return appropriate error code instead of ENOMEM

When mlx5_ib_stage_init_init() fails, return the error code related to
failure instead of -ENOMEM.

Fixes: 16c1975f1032 ("IB/mlx5: Create profile infrastructure to add and remove stages")
Link: https://lore.kernel.org/r/20210127150010.1876121-8-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/efa: Remove unused syndrome enum values

The notification syndrome enum values are unused, remove them.

Link: https://lore.kernel.org/r/20210126120702.9807-6-galpress@amazon.com
Reviewed-by: Firas JahJah <firasj@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/efa: Remove unused 'select' field from get/set feature command descriptor

The 'select' field in the get/set feature admin command is unimplemented,
unused and misleading, remove it.  The command always refers to the
current values.

Link: https://lore.kernel.org/r/20210126120702.9807-5-galpress@amazon.com
Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/efa: Remove unnecessary indentation in defs comments

The indentation in the subsequent comment lines is unnecessary, remove it.

Link: https://lore.kernel.org/r/20210126120702.9807-4-galpress@amazon.com
Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/efa: Remove duplication of upper/lower_32_bits

The EFA_DMA_ADDR_TO_UINT32_HIGH/LOW macros are the same as the kernel's
upper/lower_32_bits, remove them and use kernel macros instead.

Link: https://lore.kernel.org/r/20210126120702.9807-3-galpress@amazon.com
Reviewed-by: Firas JahJah <firasj@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/efa: Remove redundant NULL pointer check of CQE

A pointer to store the command completion must be provided as it is always
used in efa_com_put_comp_ctx() to return the completion context back to
the pool. Remove the NULL pointer check and the redundant 'status' field
stored on the context as it could be retrieved from the completion itself.

Link: https://lore.kernel.org/r/20210126120702.9807-2-galpress@amazon.com
Reviewed-by: Firas JahJah <firasj@amazon.com>
Reviewed-by: Yossi Leybovich <sleybo@amazon.com>
Signed-off-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/rc: Demote incorrectly populated kernel-doc header

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/rc.c:1388: warning: Function parameter or member 'packet' not described in 'hfi1_send_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:1388: warning: Function parameter or member 'is_fecn' not described in 'hfi1_send_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:1388: warning: Excess function parameter 'qp' description in 'hfi1_send_rc_ack'

Link: https://lore.kernel.org/r/20210126124732.3320971-21-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/verbs: Demote non-conforming doc header and fix a misspelling

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/verbs.c:741: warning: Function parameter or member 'qp' not described in 'update_tx_opstats'
 drivers/infiniband/hw/hfi1/verbs.c:1160: warning: Function parameter or member 'pkey' not described in 'egress_pkey_check'
 drivers/infiniband/hw/hfi1/verbs.c:1160: warning: Excess function parameter 'bkey' description in 'egress_pkey_check'
 drivers/infiniband/hw/hfi1/verbs.c:1217: warning: Function parameter or member 'qp' not described in 'get_send_routine'
 drivers/infiniband/hw/hfi1/verbs.c:1217: warning: Function parameter or member 'ps' not described in 'get_send_routine'

Link: https://lore.kernel.org/r/20210126124732.3320971-20-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/user_exp_rcv: Demote half-documented and kernel-doc abuses

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/user_exp_rcv.c:174: warning: Function parameter or member 'fd' not described in 'unpin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:174: warning: Function parameter or member 'tidbuf' not described in 'unpin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:174: warning: Function parameter or member 'node' not described in 'unpin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:174: warning: Function parameter or member 'idx' not described in 'unpin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:174: warning: Function parameter or member 'npages' not described in 'unpin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:174: warning: Function parameter or member 'mapped' not described in 'unpin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:196: warning: Function parameter or member 'fd' not described in 'pin_rcv_pages'
 drivers/infiniband/hw/hfi1/user_exp_rcv.c:196: warning: Function parameter or member 'tidbuf' not described in 'pin_rcv_pages'

Link: https://lore.kernel.org/r/20210126124732.3320971-19-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/ud: Fix a little more doc-rot

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/ud.c:477: warning: Function parameter or member 'ps' not described in 'hfi1_make_ud_req'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Function parameter or member 'packet' not described in 'hfi1_ud_rcv'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Excess function parameter 'ibp' description in 'hfi1_ud_rcv'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Excess function parameter 'hdr' description in 'hfi1_ud_rcv'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Excess function parameter 'rcv_flags' description in 'hfi1_ud_rcv'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Excess function parameter 'data' description in 'hfi1_ud_rcv'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Excess function parameter 'tlen' description in 'hfi1_ud_rcv'
 drivers/infiniband/hw/hfi1/ud.c:855: warning: Excess function parameter 'qp' description in 'hfi1_ud_rcv'

Link: https://lore.kernel.org/r/20210126124732.3320971-18-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/uc: Fix a little doc-rot

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/uc.c:64: warning: Function parameter or member 'ps' not described in 'hfi1_make_uc_req'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Function parameter or member 'packet' not described in 'hfi1_uc_rcv'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Excess function parameter 'ibp' description in 'hfi1_uc_rcv'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Excess function parameter 'hdr' description in 'hfi1_uc_rcv'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Excess function parameter 'rcv_flags' description in 'hfi1_uc_rcv'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Excess function parameter 'data' description in 'hfi1_uc_rcv'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Excess function parameter 'tlen' description in 'hfi1_uc_rcv'
 drivers/infiniband/hw/hfi1/uc.c:306: warning: Excess function parameter 'qp' description in 'hfi1_uc_rcv'

Link: https://lore.kernel.org/r/20210126124732.3320971-17-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/tid_rdma: Fix a plethora of kernel-doc issues

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/tid_rdma.c:321: warning: Function parameter or member 'rdi' not described in 'qp_to_rcd'
 drivers/infiniband/hw/hfi1/tid_rdma.c:321: warning: Function parameter or member 'qp' not described in 'qp_to_rcd'
 drivers/infiniband/hw/hfi1/tid_rdma.c:505: warning: Function parameter or member 'queue' not described in 'kernel_tid_waiters'
 drivers/infiniband/hw/hfi1/tid_rdma.c:536: warning: Function parameter or member 'rcd' not described in 'dequeue_tid_waiter'
 drivers/infiniband/hw/hfi1/tid_rdma.c:536: warning: Function parameter or member 'queue' not described in 'dequeue_tid_waiter'
 drivers/infiniband/hw/hfi1/tid_rdma.c:536: warning: Function parameter or member 'qp' not described in 'dequeue_tid_waiter'
 drivers/infiniband/hw/hfi1/tid_rdma.c:562: warning: Function parameter or member 'queue' not described in 'queue_qp_for_tid_wait'
 drivers/infiniband/hw/hfi1/tid_rdma.c:607: warning: Function parameter or member 'qp' not described in 'tid_rdma_schedule_tid_wakeup'
 drivers/infiniband/hw/hfi1/tid_rdma.c:639: warning: Function parameter or member 'work' not described in 'tid_rdma_trigger_resume'
 drivers/infiniband/hw/hfi1/tid_rdma.c:666: warning: Function parameter or member 'qp' not described in '_tid_rdma_flush_wait'
 drivers/infiniband/hw/hfi1/tid_rdma.c:666: warning: Function parameter or member 'queue' not described in '_tid_rdma_flush_wait'
 drivers/infiniband/hw/hfi1/tid_rdma.c:713: warning: Function parameter or member 'rcd' not described in 'kern_reserve_flow'
 drivers/infiniband/hw/hfi1/tid_rdma.c:713: warning: Function parameter or member 'last' not described in 'kern_reserve_flow'
 drivers/infiniband/hw/hfi1/tid_rdma.c:879: warning: Function parameter or member 'flow' not described in 'tid_rdma_find_phys_blocks_4k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:879: warning: Function parameter or member 'pages' not described in 'tid_rdma_find_phys_blocks_4k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:879: warning: Function parameter or member 'npages' not described in 'tid_rdma_find_phys_blocks_4k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:879: warning: Function parameter or member 'list' not described in 'tid_rdma_find_phys_blocks_4k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:972: warning: Function parameter or member 'list' not described in 'tid_flush_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:972: warning: Function parameter or member 'idx' not described in 'tid_flush_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:972: warning: Function parameter or member 'pages' not described in 'tid_flush_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:972: warning: Function parameter or member 'sets' not described in 'tid_flush_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1017: warning: Function parameter or member 'flow' not described in 'tid_rdma_find_phys_blocks_8k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1017: warning: Function parameter or member 'pages' not described in 'tid_rdma_find_phys_blocks_8k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1017: warning: Function parameter or member 'npages' not described in 'tid_rdma_find_phys_blocks_8k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1017: warning: Function parameter or member 'list' not described in 'tid_rdma_find_phys_blocks_8k'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1083: warning: Function parameter or member 'flow' not described in 'kern_find_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1083: warning: Function parameter or member 'pages' not described in 'kern_find_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1083: warning: Function parameter or member 'ss' not described in 'kern_find_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1083: warning: Function parameter or member 'last' not described in 'kern_find_pages'
 drivers/infiniband/hw/hfi1/tid_rdma.c:1604: warning: Function parameter or member 'req' not described in 'hfi1_kern_exp_rcv_free_flows'
 drivers/infiniband/hw/hfi1/tid_rdma.c:3458: warning: Function parameter or member 'qp' not described in 'hfi1_tid_write_alloc_resources'
 drivers/infiniband/hw/hfi1/tid_rdma.c:3458: warning: Function parameter or member 'intr_ctx' not described in 'hfi1_tid_write_alloc_resources'

Link: https://lore.kernel.org/r/20210126124732.3320971-16-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/sdma: Fix misnaming of 'sdma_send_txlist()'s 'count_out' param

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/sdma.c:2476: warning: Function parameter or member 'count_out' not described in 'sdma_send_txlist'
 drivers/infiniband/hw/hfi1/sdma.c:2476: warning: Excess function parameter 'count' description in 'sdma_send_txlist'

Link: https://lore.kernel.org/r/20210126124732.3320971-15-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/ruc: Fix a small formatting and description issues

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/ruc.c:277: warning: Function parameter or member 'bth1' not described in 'hfi1_make_ruc_header_16B'
 drivers/infiniband/hw/hfi1/ruc.c:365: warning: Function parameter or member 'bth1' not described in 'hfi1_make_ruc_header_9B'
 drivers/infiniband/hw/hfi1/ruc.c:472: warning: Function parameter or member 'tid' not described in 'hfi1_schedule_send_yield'
 drivers/infiniband/hw/hfi1/ruc.c:472: warning: Excess function parameter 'timeout' description in 'hfi1_schedule_send_yield'

Link: https://lore.kernel.org/r/20210126124732.3320971-14-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/qp: Fix some formatting issues and demote kernel-doc abuse

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/qp.c:195: warning: Function parameter or member 'dev' not described in 'verbs_mtu_enum_to_int'
 drivers/infiniband/hw/hfi1/qp.c:195: warning: Function parameter or member 'mtu' not described in 'verbs_mtu_enum_to_int'
 drivers/infiniband/hw/hfi1/qp.c:306: warning: Function parameter or member 'qp' not described in 'hfi1_setup_wqe'
 drivers/infiniband/hw/hfi1/qp.c:306: warning: Function parameter or member 'wqe' not described in 'hfi1_setup_wqe'
 drivers/infiniband/hw/hfi1/qp.c:306: warning: Function parameter or member 'call_send' not described in 'hfi1_setup_wqe'
 drivers/infiniband/hw/hfi1/qp.c:922: warning: Function parameter or member 'qp' not described in 'hfi1_qp_iter_cb'
 drivers/infiniband/hw/hfi1/qp.c:922: warning: Function parameter or member 'v' not described in 'hfi1_qp_iter_cb'

Link: https://lore.kernel.org/r/20210126124732.3320971-13-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/rc: Fix a few function documentation issues

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/rc.c:430: warning: Function parameter or member 'ps' not described in 'hfi1_make_rc_req'
 drivers/infiniband/hw/hfi1/rc.c:1387: warning: Function parameter or member 'packet' not described in 'hfi1_send_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:1387: warning: Function parameter or member 'is_fecn' not described in 'hfi1_send_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:1387: warning: Excess function parameter 'qp' description in 'hfi1_send_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:2008: warning: Function parameter or member 'aeth' not described in 'do_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:2008: warning: Function parameter or member 'val' not described in 'do_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:2008: warning: Function parameter or member 'rcd' not described in 'do_rc_ack'
 drivers/infiniband/hw/hfi1/rc.c:2554: warning: Function parameter or member 'rcd' not described in 'rc_rcv_error'

Link: https://lore.kernel.org/r/20210126124732.3320971-12-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/pio_copy: Provide entry for 'pio_copy()'s 'dd' param

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/pio_copy.c:73: warning: Function parameter or member 'dd' not described in 'pio_copy'

Link: https://lore.kernel.org/r/20210126124732.3320971-11-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/pcie: Demote kernel-doc abuses

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/pcie.c:343: warning: Function parameter or member 'dd' not described in 'restore_pci_variables'
 drivers/infiniband/hw/hfi1/pcie.c:402: warning: Function parameter or member 'dd' not described in 'save_pci_variables'

Link: https://lore.kernel.org/r/20210126124732.3320971-10-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/netdev_rx: Fix misdocumentation of the 'start_id' param

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/netdev_rx.c:473: warning: Function parameter or member 'start_id' not described in 'hfi1_netdev_get_first_data'
 drivers/infiniband/hw/hfi1/netdev_rx.c:473: warning: Excess function parameter 'id' description in 'hfi1_netdev_get_first_data'

Link: https://lore.kernel.org/r/20210126124732.3320971-9-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/sw/rdmavt/qp: Fix kernel-doc formatting problem

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/sw/rdmavt/qp.c:1929: warning: Function parameter or member 'post_parms' not described in 'rvt_qp_valid_operation'

Link: https://lore.kernel.org/r/20210126124732.3320971-8-lee.jones@linaro.org
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/sw/rdmavt/mad: Fix misspelling of 'rvt_process_mad()'s 'in_mad_size' param

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/sw/rdmavt/mad.c:78: warning: Function parameter or member 'in_mad_size' not described in 'rvt_process_mad'

Link: https://lore.kernel.org/r/20210126124732.3320971-7-lee.jones@linaro.org
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/msix: Add description for 'name' and remove superfluous param 'idx'

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/msix.c:120: warning: Function parameter or member 'name' not described in 'msix_request_irq'
 drivers/infiniband/hw/hfi1/msix.c:120: warning: Excess function parameter 'idx' description in 'msix_request_irq'

Link: https://lore.kernel.org/r/20210126124732.3320971-6-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/mad: Demote half-completed kernel-doc header fix another

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/mad.c:1354: warning: Function parameter or member 'am' not described in '__subn_set_opa_portinfo'
 drivers/infiniband/hw/hfi1/mad.c:1354: warning: Function parameter or member 'data' not described in '__subn_set_opa_portinfo'
 drivers/infiniband/hw/hfi1/mad.c:1354: warning: Function parameter or member 'resp_len' not described in '__subn_set_opa_portinfo'
 drivers/infiniband/hw/hfi1/mad.c:1354: warning: Function parameter or member 'max_len' not described in '__subn_set_opa_portinfo'
 drivers/infiniband/hw/hfi1/mad.c:1354: warning: Function parameter or member 'local_mad' not described in '__subn_set_opa_portinfo'
 drivers/infiniband/hw/hfi1/mad.c:4919: warning: Function parameter or member 'out_mad_size' not described in 'hfi1_process_mad'
 drivers/infiniband/hw/hfi1/mad.c:4919: warning: Function parameter or member 'out_mad_pkey_index' not described in 'hfi1_process_mad'

Link: https://lore.kernel.org/r/20210126124732.3320971-5-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/iowait: Demote half-completed kernel-doc and fix formatting issue in another

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/iowait.c:52: warning: Function parameter or member 'tidfunc' not described in 'iowait_init'
 drivers/infiniband/hw/hfi1/iowait.c:52: warning: Function parameter or member 'wakeup' not described in 'iowait_init'
 drivers/infiniband/hw/hfi1/iowait.c:52: warning: Function parameter or member 'sdma_drained' not described in 'iowait_init'
 drivers/infiniband/hw/hfi1/iowait.c:52: warning: Function parameter or member 'init_priority' not described in 'iowait_init'
 drivers/infiniband/hw/hfi1/iowait.c:52: warning: Excess function parameter 'resume' description in 'iowait_init'
 drivers/infiniband/hw/hfi1/iowait.c:94: warning: Function parameter or member 'w' not described in 'iowait_set_work_flag'

Link: https://lore.kernel.org/r/20210126124732.3320971-4-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/sw/rdmavt/srq: Fix a couple of kernel-doc issues

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/sw/rdmavt/srq.c:78: warning: Function parameter or member 'ibsrq' not described in 'rvt_create_srq'
 drivers/infiniband/sw/rdmavt/srq.c:78: warning: Excess function parameter 'ibpd' description in 'rvt_create_srq'
 drivers/infiniband/sw/rdmavt/srq.c:336: warning: Function parameter or member 'udata' not described in 'rvt_destroy_srq'

Link: https://lore.kernel.org/r/20210126124732.3320971-3-lee.jones@linaro.org
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hw/hfi1/intr: Fix some kernel-doc formatting issues

Fixes the following W=1 kernel build warning(s):

 drivers/infiniband/hw/hfi1/intr.c:99: warning: Function parameter or member 'msg' not described in 'format_hwmsg'
 drivers/infiniband/hw/hfi1/intr.c:99: warning: Function parameter or member 'msgl' not described in 'format_hwmsg'
 drivers/infiniband/hw/hfi1/intr.c:99: warning: Function parameter or member 'hwmsg' not described in 'format_hwmsg'
 drivers/infiniband/hw/hfi1/intr.c:115: warning: Function parameter or member 'hwerrs' not described in 'hfi1_format_hwerrors'
 drivers/infiniband/hw/hfi1/intr.c:115: warning: Function parameter or member 'hwerrmsgs' not described in 'hfi1_format_hwerrors'
 drivers/infiniband/hw/hfi1/intr.c:115: warning: Function parameter or member 'nhwerrmsgs' not described in 'hfi1_format_hwerrors'
 drivers/infiniband/hw/hfi1/intr.c:115: warning: Function parameter or member 'msg' not described in 'hfi1_format_hwerrors'
 drivers/infiniband/hw/hfi1/intr.c:115: warning: Function parameter or member 'msgl' not described in 'hfi1_format_hwerrors'

Link: https://lore.kernel.org/r/20210126124732.3320971-2-lee.jones@linaro.org
Cc: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Cc: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Replace missing rxe_pool_get_index_locked

One of the pool APIs for when caller is holding lock was not defined but
is declared in rxe_pool.h. This patch adds the definition.

Link: https://lore.kernel.org/r/20210125211641.2694-7-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Remove unneeded pool->state

rxe_pool.c uses the field pool->state to mark a pool as invalid when it is
shut down and checks it in several pool APIs to verify that the pool has
not been shut down. This is unneeded because the pools are not marked
invalid unless the entire driver is being removed at which point no
functional APIs should or could be executing. This patch removes this
field and associated code.

Link: https://lore.kernel.org/r/20210125211641.2694-6-rpearson@hpe.com
Suggested-by: zyjzyj2000@gmail.c
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Remove references to ib_device and pool

rxe_pool.c takes references to the pool and ib_device structs for each
object allocated and also keeps an atomic num_elem count in each
pool. This is more work than is needed. Pool allocation is only called
from verbs APIs which already have references to ib_device and pools are
only diasbled when the driver is removed so no protection of the pool
addresses are needed. The elem count is used to warn if elements are still
present in a pool when it is cleaned up which is useful.

This patch eliminates the references to the ib_device and pool structs.

Link: https://lore.kernel.org/r/20210125211641.2694-5-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Remove RXE_POOL_ATOMIC

rxe_alloc() used the RXE_POOL_ATOMIC flag in rxe_type_info to select
GFP_ATOMIC in calls to kzalloc(). This was intended to handle cases where
an object could be created in interrupt context. This no longer occurs
since allocating those objects has moved into the core so this flag is not
necessary. An incorrect use of this flag was still present for rxe_mc_elem
objects and is removed.

Link: https://lore.kernel.org/r/20210125211641.2694-4-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix misleading comments and names

The names and comments of the 'unlocked' pool APIs are very misleading and
not what was intended. This patch replaces 'rxe_xxx_nl' with
'rxe_xxx_locked' with comments indicating that the caller is expected to
hold the rxe pool lock.

Link: https://lore.kernel.org/r/20210125211641.2694-3-rpearson@hpe.com
Reported-by: Hillf Danton <hdanton@sina.com>
Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix bug in rxe_alloc()

A recent patch which added an 'unlocked' version of rxe_alloc introduced a
bug causing kzalloc(..., GFP_KERNEL) to be called while holding a spin
lock. This patch corrects that error.

rxe_alloc_nl() should always be called while holding the pool->pool_lock
so the 2nd argument to kzalloc there should be GFP_ATOMIC.

rxe_alloc() prior to the change only locked the code around checking that
pool->state is RXE_POOL_STATE_VALID to avoid races between working threads
and a thread shutting down the rxe driver. This patch reverts rxe_alloc()
to this behavior so the lock is not held when kzalloc() is called.

Link: https://lore.kernel.org/r/20210125211641.2694-2-rpearson@hpe.com
Reported-by: syzbot+ec2fd72374785d0e558e@syzkaller.appspotmail.com
Fixes: 3853c35e243d ("RDMA/rxe: Add unlocked versions of pool APIs")
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/umad: Return EPOLLERR in case of when device disassociated

Currently, polling a umad device will always works, even if the device was
disassociated. A disassociated device should immediately return EPOLLERR
from poll(). Otherwise userspace is endlessly hung on poll() with no idea
that the device has been removed from the system.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Link: https://lore.kernel.org/r/20210125121339.837518-3-leon@kernel.org
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/umad: Return EIO in case of when device disassociated

MAD message received by the user has EINVAL error in all flows
including when the device is disassociated. That makes it impossible
for the applications to treat such flow differently.

Change it to return EIO, so the applications will be able to perform
disassociation recovery.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Link: https://lore.kernel.org/r/20210125121339.837518-2-leon@kernel.org
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>