IB/mlx5: Fix Raw Packet QP event handler assignment

In case we have SQ and RQ for Raw Packet QP, the SQ's event handler
wasn't assigned.

Fixing this by assigning event handler for each WQ after creation.

[ 1877.145243] Call Trace:
[ 1877.148644] <IRQ>
[ 1877.150580] [<ffffffffa07987c5>] ? mlx5_rsc_event+0x105/0x210 [mlx5_core]
[ 1877.159581] [<ffffffffa0795bd7>] ? mlx5_cq_event+0x57/0xd0 [mlx5_core]
[ 1877.167137] [<ffffffffa079208e>] mlx5_eq_int+0x53e/0x6c0 [mlx5_core]
[ 1877.174526] [<ffffffff8101a679>] ? sched_clock+0x9/0x10
[ 1877.180753] [<ffffffff810f717e>] handle_irq_event_percpu+0x3e/0x1e0
[ 1877.188014] [<ffffffff810f735d>] handle_irq_event+0x3d/0x60
[ 1877.194567] [<ffffffff810f9fe7>] handle_edge_irq+0x77/0x130
[ 1877.201129] [<ffffffff81014c3f>] handle_irq+0xbf/0x150
[ 1877.207244] [<ffffffff815ed78a>] ? atomic_notifier_call_chain+0x1a/0x20
[ 1877.214829] [<ffffffff815f434f>] do_IRQ+0x4f/0xf0
[ 1877.220498] [<ffffffff815e94ad>] common_interrupt+0x6d/0x6d
[ 1877.227025] <EOI>
[ 1877.228967] [<ffffffff814834e2>] ? cpuidle_enter_state+0x52/0xc0
[ 1877.236990] [<ffffffff81483615>] cpuidle_idle_call+0xc5/0x200
[ 1877.243676] [<ffffffff8101bc7e>] arch_cpu_idle+0xe/0x30
[ 1877.249831] [<ffffffff810b4725>] cpu_startup_entry+0xf5/0x290
[ 1877.256513] [<ffffffff815cfee1>] start_secondary+0x265/0x27b
[ 1877.263111] Code: Bad RIP value.
[ 1877.267296] RIP [< (null)>] (null)
[ 1877.273264] RSP <ffff88046fd63df8>
[ 1877.277531] CR2: 0000000000000000

Fixes: 19098df2da78 ("IB/mlx5: Refactor mlx5_ib_qp to accommodate other QP types")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Avoid accessing non-allocated memory when inferring port type

Commit 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
introduced the concept of type in ah_attr:
 * During ib_register_device, each port is checked for its type which
   is stored in ib_device's port_immutable array.
 * During uverbs' modify_qp, the type is inferred using the port number
   in ib_uverbs_qp_dest struct (address vector) by accessing the
   relevant port_immutable array and the type is passed on to
   providers.

IB spec (version 1.3) enforces a valid port value only in Reset to
Init. During Init to RTR, the address vector must be valid but port
number is not mentioned as a field in the address vector, so its
value is not validated, which leads to accesses to a non-allocated
memory when inferring the port type.

Save the real port number in ib_qp during modify to Init (when the
comp_mask indicates that the port number is valid) and use this value
to infer the port type.

Avoid copying the address vector fields if the matching bit is not set
in the attr_mask. Address vector can't be modified before the port, so
no valid flow is affected.

Fixes: 44c58487d51a ('IB/core: Define 'ib' and 'roce' rdma_ah_attr types')
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/uverbs: Initialize cq_context appropriately

Initializing cq_context with ev_queue in create_cq(), leads to NULL pointer
dereference in ib_uverbs_comp_handler(), if application doesnot use completion
channel. This patch fixes the cq_context initialization.

Fixes: 1e7710f3f65 ("IB/core: Change completion channel to use the reworked")
Cc: stable@vger.kernel.org # 4.12
Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
(cherry picked from commit 699a2d5b1b880b4e4e1c7d55fa25659322cf5b51)

IB/uverbs: Fix NULL pointer dereference during device removal

As part of ib_uverbs_remove_one which might be triggered upon
reset flow, we trigger IB_EVENT_DEVICE_FATAL event to userspace
application.
If device was removed after uverbs fd was opened but before
ib_uverbs_get_context was called, the event file will be accessed
before it was allocated, result in NULL pointer dereference:

[ 72.325873] BUG: unable to handle kernel NULL pointer dereference at (null)
...
[ 72.325984] IP: _raw_spin_lock_irqsave+0x22/0x40
[ 72.327123] Call Trace:
[ 72.327168] ib_uverbs_async_handler.isra.8+0x2e/0x160 [ib_uverbs]
[ 72.327216] ? synchronize_srcu_expedited+0x27/0x30
[ 72.327269] ib_uverbs_remove_one+0x120/0x2c0 [ib_uverbs]
[ 72.327330] ib_unregister_device+0xd0/0x180 [ib_core]
[ 72.327373] mlx5_ib_remove+0x74/0x140 [mlx5_ib]
[ 72.327422] mlx5_remove_device+0xfb/0x110 [mlx5_core]
[ 72.327466] mlx5_unregister_interface+0x3c/0xa0 [mlx5_core]
[ 72.327509] mlx5_ib_cleanup+0x10/0x962 [mlx5_ib]
[ 72.327546] SyS_delete_module+0x155/0x230
[ 72.328472] ? exit_to_usermode_loop+0x70/0xa6
[ 72.329370] do_syscall_64+0x54/0xc0
[ 72.330262] entry_SYSCALL64_slow_path+0x25/0x25

Fix it by checking that user context was allocated before
trigger the event.

Fixes: 036b10635739 ('IB/uverbs: Enable device removal when there are active user space applications')
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Protect sysfs entry on ib_unregister_device

ib_unregister_device is not protecting removal of sysfs entries.
A call to ib_register_device in that window can result in
duplicate sysfs entry warning. Move mutex_unlock to after
ib_device_unregister_sysfs to protect against sysfs entry creation.

This issue is exposed during driver load/unload stress test.

WARNING: CPU: 5 PID: 4445 at fs/sysfs/dir.c:31 sysfs_warn_dup+0x5f/0x70
sysfs: cannot create duplicate filename '/class/infiniband/i40iw0'
Hardware name: Gigabyte Technology Co., Ltd. To be filled by O.E.M./Q87M-D2H
BIOS F7 01/17/2014
Workqueue: i40e i40e_service_task [i40e]
Call Trace:
dump_stack+0x67/0x98
__warn+0xcc/0xf0
warn_slowpath_fmt+0x4a/0x50
? kernfs_path_from_node+0x4b/0x60
sysfs_warn_dup+0x5f/0x70
sysfs_do_create_link_sd.isra.2+0xb7/0xc0
sysfs_create_link+0x20/0x40
device_add+0x28c/0x600
ib_device_register_sysfs+0x58/0x170 [ib_core]
ib_register_device+0x325/0x570 [ib_core]
? i40iw_register_rdma_device+0x1f4/0x400 [i40iw]
? kmem_cache_alloc_trace+0x143/0x330
? __raw_spin_lock_init+0x2d/0x50
i40iw_register_rdma_device+0x2dc/0x400 [i40iw]
i40iw_open+0x10a6/0x1950 [i40iw]
? i40iw_open+0xeab/0x1950 [i40iw]
? i40iw_make_cm_node+0x9c0/0x9c0 [i40iw]
i40e_client_subtask+0xa4/0x110 [i40e]
i40e_service_task+0xc2d/0x1320 [i40e]
process_one_work+0x203/0x710
? process_one_work+0x16f/0x710
worker_thread+0x126/0x4a0
? trace_hardirqs_on+0xd/0x10
kthread+0x112/0x150
? process_one_work+0x710/0x710
? kthread_create_on_node+0x40/0x40
ret_from_fork+0x2e/0x40
---[ end trace fd11b69e21ea7653 ]---
Couldn't register device i40iw0 with driver model

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Sindhu Devale <sindhu.devale@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

iw_cxgb4: fix misuse of integer variable

Fixes: ee30f7d507c0 ("iw_cxgb4: Max fastreg depth depends on DSGL support")
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hns: fix memory leak on ah on error return path

When dmac is NULL, ah is not being freed on the error return path. Fix
this by kfree'ing it.

Detected by CoverityScan, CID#1452636 ("Resource Leak")

Fixes: d8966fcd4c25 ("IB/core: Use rdma_ah_attr accessor functions")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Fix potential fcn_id_array out of bounds

Avoid out of bounds error by utilizing I40IW_MAX_STATS_COUNT
instead of I40IW_INVALID_FCN_ID.

Signed-off-by: Christopher N Bednarz <christoper.n.bednarz@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Use correct alignment for CQ0 memory

Utilize correct alignment variable when allocating
DMA memory for CQ0.

Signed-off-by: Christopher N Bednarz <christopher.n.bednarz@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Fix typecast of tcp_seq_num

The typecast of tcp_seq_num incorrectly uses u8. Fix by
casting to u32.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Correct variable names

Fix incorrect naming of status code and struct. Use inline
instead of immediate.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Fix parsing of query/commit FPM buffers

Parsing of commit/query Host Memory Cache Function Private Memory
is not skipping over reserved fields and incorrectly assigning
those values into object's base/cnt/max_cnt fields. Skip over
reserved fields and set correct values. Also correct memory
alignment requirement for commit/query FPM buffers.

Signed-off-by: Chien Tin Tung <chien.tin.tung@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Christopher N Bednarz <christopher.n.bednarz@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/vmw_pvrdma: Report CQ missed events

There is a chance of a race between arming the CQ and receiving
completions. By reporting CQ missed events any ULPs should poll
again to get the completions.

Fixes: 29c8d9eba550 ("IB: Add vmw_pvrdma driver")
Acked-by: Aditya Sarwade <asarwade@vmware.com>
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge tag 'rdma-rc-2017-07-26' of git://git./linux/kernel/git/leon/linux-rdma into leon-ipoib

IPoIB fixes for 4.13

The patchset provides various fixes for IPoIB. It is combination of
fixes to various issues discovered during verification along with
static checkers cleanup patches.

Most of the patches are from pre-git era and hence lack of Fixes lines.

There is one exception in this IPoIB group - addition of patch revert:
Revert "IB/core: Allow QP state transition from reset to error", but
it followed by proper fix to the annoying print, so I thought it is
appropriate to include it.

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

IB/hns: checking for IS_ERR() instead of NULL

The hns_roce_v1_create_lp_qp() returns NULL on error, not error pointers.

Fixes: bfcc681bd09d ("IB/hns: Fix the bug when free mr")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/mlx5: Fix existence check for extended address vector

The extended address vector is the highest bit in be32 variable,
but it was compared with the lowest. This patch fixes the endianness
of that check and removes already declared define.

Fixes: 17d2f88f92ce ("IB/mlx5: Add ODP atomics support")
Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/uverbs: Fix device cleanup

Uverbs device should be cleaned up only when there is no
potential usage of.

As part of ib_uverbs_remove_one which might be triggered upon reset flow
the device reference count is decreased as expected and leave the final
cleanup to the FDs that were opened.

Current code increases reference count upon opening a new command FD and
decreases it upon closing the file. The event FD is opened internally
and rely on the command FD by taking on it a reference count.

In case that the command FD was closed and just later the event FD we
may ensure that the device resources as of srcu are still alive as they
are still in use.

Fixing the above by moving the reference count decreasing to the place
where the command FD is really freed instead of doing that when it was
just closed.

fixes: 036b10635739 ("IB/uverbs: Enable device removal when there are active user space applications")
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Tested-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/uverbs: Prevent leak of reserved field

initialize to zero the response structure to prevent
the leakage of "resp.reserved" field.

drivers/infiniband/core/uverbs_cmd.c:1178 ib_uverbs_resize_cq() warn:
check that 'resp.reserved' doesn't leak information

Fixes: 33b9b3ee9709 ("IB: Add userspace support for resizing CQs")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Fix race condition in resolving IP to MAC

Currently while resolving IP address to MAC address single delayed work
is used for resolving multiple such resolve requests. This singled work
is essentially performs two tasks.
(a) any retry needed to resolve and
(b) it executes the callback function for all completed requests

While work is executing callbacks, any new work scheduled on for this
workqueue is lost because workqueue has completed looking at all pending
requests and now looking at callbacks, but work is still under
execution. Any further retry to look at pending requests in
process_req() after executing callbacks would lead to similar race
condition (may be reduce the probably further but doesn't eliminate it).
Retrying to enqueue work that from queue_req() context is not something
rest of the kernel modules have followed.

Therefore fix in this patch utilizes kernel facility to enqueue multiple
work items to a workqueue. This ensures that no such requests
gets lost in synchronization. Request list is still maintained so that
rdma_cancel_addr() can unlink the request and get the completion with
error sooner. Neighbour update event handling continues to be handled in
same way as before.
Additionally process_req() work entry cancels any pending work for a
request that gets completed while processing those requests.

Originally ib_addr was ST workqueue, but it became MT work queue with
patch of [1]. This patch again makes it similar to ST so that
neighbour update events handler work item doesn't race with
other work items.

In one such below trace, (though on 4.5 based kernel) it can be seen
that process_req() never executed the callback, which is likely for an
event that was schedule by queue_req() when previous callback was
getting executed by workqueue.

 [<ffffffff816b0dde>] schedule+0x3e/0x90
 [<ffffffff816b3c45>] schedule_timeout+0x1b5/0x210
 [<ffffffff81618c37>] ? ip_route_output_flow+0x27/0x70
 [<ffffffffa027f9c9>] ? addr_resolve+0x149/0x1b0 [ib_addr]
 [<ffffffff816b228f>] wait_for_completion+0x10f/0x170
 [<ffffffff810b6140>] ? try_to_wake_up+0x210/0x210
 [<ffffffffa027f220>] ? rdma_copy_addr+0xa0/0xa0 [ib_addr]
 [<ffffffffa0280120>] rdma_addr_find_l2_eth_by_grh+0x1d0/0x278 [ib_addr]
 [<ffffffff81321297>] ? sub_alloc+0x77/0x1c0
 [<ffffffffa02943b7>] ib_init_ah_from_wc+0x3a7/0x5a0 [ib_core]
 [<ffffffffa0457aba>] cm_req_handler+0xea/0x580 [ib_cm]
 [<ffffffff81015982>] ? __switch_to+0x212/0x5e0
 [<ffffffffa04582fd>] cm_work_handler+0x6d/0x150 [ib_cm]
 [<ffffffff810a14c1>] process_one_work+0x151/0x4b0
 [<ffffffff810a1940>] worker_thread+0x120/0x480
 [<ffffffff816b074b>] ? __schedule+0x30b/0x890
 [<ffffffff810a1820>] ? process_one_work+0x4b0/0x4b0
 [<ffffffff810a1820>] ? process_one_work+0x4b0/0x4b0
 [<ffffffff810a6b1e>] kthread+0xce/0xf0
 [<ffffffff810a6a50>] ? kthread_freezable_should_stop+0x70/0x70
 [<ffffffff816b53a2>] ret_from_fork+0x42/0x70
 [<ffffffff810a6a50>] ? kthread_freezable_should_stop+0x70/0x70
INFO: task kworker/u144:1:156520 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this
message.
kworker/u144:1  D ffff883ffe1d7600     0 156520      2 0x00000080
Workqueue: ib_addr process_req [ib_addr]
 ffff883f446fbbd8 0000000000000046 ffff881f95280000 ffff881ff24de200
 ffff883f66120000 ffff883f446f8008 ffff881f95280000 ffff883f6f9208c4
 ffff883f6f9208c8 00000000ffffffff ffff883f446fbbf8 ffffffff816b0dde

[1] http://lkml.iu.edu/hypermail/linux/kernel/1608.1/05834.html

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/ipoib: Notify on modify QP failure only when relevant

Modify QP can fail and it can be acceptable, like when moving from RST to
ERR state, all the rest are not acceptable and a message to the log
should be printed.

The current code prints on all failures and many messages like:
"Failed to modify QP to ERROR state" appear, even when supported by the
state machine of the QP object.

Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

Revert "IB/core: Allow QP state transition from reset to error"

The commit ebc9ca43e1d5 ("IB/core: Allow QP state transition from reset to error")
allowed transition from Reset to Error state for the QPs. This behavior
doesn't follow the IBTA specification 1.3, which in 10.3.1 QUEUE PAIR AND
EE CONTEXT STATES section.

The quote from the spec:
"An error can be forced from any state, except Reset, with
the Modify QP/EE Verb."

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

IB/ipoib: Remove double pointer assigning

There is no need to assign "p" pointer twice.

This patch fixes the following smatch warning:
drivers/infiniband/ulp/ipoib/ipoib_cm.c:517 ipoib_cm_rx_handler() warn:
missing break? reassigning 'p->id'

Fixes: 839fcaba355a ("IPoIB: Connected mode experimental support")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

IB/ipoib: Clean error paths in add port

Refactor error paths in ipoib_add_port() function. The code flow
ensures that the function terminates on every error flow and it makes
redundant all "else" cases.

The functions are called during the flow are returning "result < 0", in
case of error, so there is no need to check it explicitly.

Fixes: 58e9cc90cda7 ("IB/IPoIB: Fix bad error flow in ipoib_add_port()")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

IB/ipoib: Add get statistics support to SRIOV VF

Add SRIOV VF support to get traffic statistics.

Signed-off-by: Feras Daoud <ferasda@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

IB/ipoib: Add multicast packets statistics

Update the multicast counter when multicast packets are received and
provide this information through ethtool support.

Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

IB/ipoib: Set IPOIB_NEIGH_TBL_FLUSH after flushed completion initialization

Set IPOIB_NEIGH_TBL_FLUSH bit after initializing the neighbor
flushed completion, otherwise the garbage collector may signal
a completion while it is not initialized yet.

Fixes: b63b70d87741 ("IPoIB: Use a private hash table for path lookup in xmit path")
Signed-off-by: Feras Daoud <ferasda@mellanox.com>
Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

IB/ipoib: Prevent setting negative values to max_nonsrq_conn_qp

Don't allow negative values to max_nonsrq_conn_qp. There is no functional
impact on a negative value but it is logicically incorrect.

Fixes: 68e995a29572 ("IPoIB/cm: Add connected mode support for devices without SRQs")
Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

IB/ipoib: Make sure no in-flight joins while leaving that mcast

While cleaning neighs and there is a send-only mcast neigh, the driver
should wait to finish its join process before trying to remove it.

Without this patch, we will see messages like: "ipoib_mcast_leave on an
in-flight join" and unexpected results in the join_complete.

Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

IB/ipoib: Use cancel_delayed_work_sync when needed

The work mcast_task can re-queue itself, so instead of doing
cancel && flush_workqueue, that still can leave a queued task
on the air, use cancel_delayed_work_sync.

Also, no need to use lock over the cancel, the original lock was
due to bit assignment setting (IPOIB_MCAST_RUN) that is not in use
anymore.

Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

IB/ipoib: Fix race between light events and interface restart

A potential race between light_event and interface restart
may attach multicast group to an already attached QP.

Scenario:
light_event flow goes through ipoib_mcast_dev_flush function,
if a context switch occurs before calling ipoib_mcast_remove_list,
then we may face a situation where the broadcast of the priv is null
and the corresponding QP is not detached yet.
If an "interface restart" runs during the previous context switch,
the following scenario occurs:
When the device goes up, ipoib_ib_dev_up function will be called,
it will send a new registration request to the broadcast group and then
attach the group to the QP that was not detached before.

     IPOIB_FLUSH_LIGHT                                          INTERFACE RESTART

    __ipoib_ib_dev_flush                                                |
        |                                                               |
        |                                                               |
        |                                                               |
    ipoib_mcast_dev_flush                                               |
    Move mcast list and broadcast to remove_list                        |
        |                                                               |
        |                                                               |
    Context Switch-->                                                   |
        |                                                       ipoib_ib_dev_down
        |                                                               |
        |                                                               |
        |                                                       ipoib_ib_dev_up
        |                                                               |
        |                                                               |
        |                                                       ipoib_mcast_join_task
        |                                                       allocate new broadcast
        |                                                               |
        |                                                               |
        |                                                       Attach QP to multicast group
        |                                                               |
        |                                                               |
        |                                                       <--Context Switch
    ipoib_mcast_leave
    Detach QP from multicast group

Signed-off-by: Feras Daoud <ferasda@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/core: Initialize port_num in qp_attr

Initialize the port_num for iWARP in rdma_init_qp_attr.

Fixes: 5ecce4c9b17b("Check port number supplied by user verbs cmds")
Cc: <stable@vger.kernel.org> # v2.6.14+
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Tested-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/uverbs: Fix the check for port number

The port number is only valid if IB_QP_PORT is set in the mask.
So only check port number if it is valid to prevent modify_qp from
failing due to an invalid port number.

Fixes: 5ecce4c9b17b("Check port number supplied by user verbs cmds")
Cc: <stable@vger.kernel.org> # v2.6.14+
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Tested-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/cma: Fix reference count leak when no ipv4 addresses are set

Once in_dev_get is called to receive in_device pointer, the
in_device reference counter is increased, but if there are
no ipv4 addresses configured on the net-device the ifa_list
will be null, resulting in a flow that doesn't call in_dev_put
to decrease the ref_cnt.
This was exposed when running RoCE over ipv6 without any ipv4
addresses configured

Fixes: commit 8e3867310c90 ("IB/cma: Fix a race condition in iboe_addr_get_sgid()")

Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/iser: don't send an rkey if all data is written as immadiate-data

We might get some bogus error completions in case the target will
remotely invalidate the rkey and the HCA will need to retransmit
from this buffer.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>

rxe: fix broken receive queue draining

If we modified the qp to ERROR state, and
drained the recieve queue, post_recv must
trigger the responder task to complete
the drain work request.

Cc: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Vijay Immanuel <vijayi@attalasystems.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>--
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/qedr: Prevent memory overrun in verbs' user responses

Wrap ib_copy_to_udata with a function that ensures that the data
being copied over to user space isn't longer than the allowed.

Fixes: cecbcddf6461 ("qedr: Add support for QP verbs")
Fixes: a7efd7773e31 ("qedr: Add support for PD,PKEY and CQ verbs")
Fixes: ac1b36e55a51 ("qedr: Add support for user context verbs")
Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

iw_cxgb4: don't use WR keys/addrs for 0 byte reads

Only use the read sge lkey/addr and the remote rkey/addr if the
length of the read is not zero. Otherwise the read response might
be treated as the RTR read response and not delivered to the
application. Or worse Terminator hardware will fail a 0B read
if the STAG is 0 even if the read length is 0.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx4: Fix CM REQ retries in paravirt mode

CM REQs cannot be successfully retried, because a new pv_cm_id is
created for each request, without checking if one already exists.

By checking if an id exists before creating one, the bug is fixed.

This bug can be provoked by running an RDMA CM user-land application,
but inserting a five seconds delay before the rdma_accept() call on
the passive side. This delay is larger than the default CMA timeout,
and triggers a retry from the active side. The retried REQ will use
another pv_cm_id (the cm_id on the wire). This confuses the CM
protocol and two REJs are sent from the passive side.

Here is an excerpt from ibdump running without the patch:

3.285092       LID: 4 -> LID: 4       SDP 290 CM: ConnectRequest(SDP Hello)
7.382711       LID: 4 -> LID: 4       SDP 290 CM: ConnectRequest(SDP Hello)
7.382861       LID: 4 -> LID: 4       InfiniBand 290 CM: ConnectReject
7.387644       LID: 4 -> LID: 4       InfiniBand 290 CM: ConnectReject

and here is the same with bug fix applied:

3.251010       LID: 4 -> LID: 4       SDP 290 CM: ConnectRequest(SDP Hello)
7.349387       LID: 4 -> LID: 4       SDP 290 CM: ConnectRequest(SDP Hello)
8.258443       LID: 4 -> LID: 4       SDP 290 CM: ConnectReply(SDP Hello)
8.259890       LID: 4 -> LID: 4       InfiniBand 290 CM: ReadyToUse

Suggested-by: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
Reported-by: Wei Lin Guay <wei.lin.guay@oracle.com>
Tested-by: Wei Lin Guay <wei.lin.guay@oracle.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Acked-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rdmavt: Setting of QP timeout can overflow jiffies computation

Current computation of qp->timeout_jiffies in rvt_modify_qp() will cause
overflow due to the fact that the input to the function usecs_to_jiffies
is only 32-bit ( unsigned int). Overflow will occur when attr->timeout is
equal to or greater than 30. The consequence is unnecessarily excessive
retry and thus degradation of the system performance.

This patch fixes the problem by limiting the input to 5-bit and calling
usecs_to_jiffies() before multiplying the scaling factor.

Reviewed-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: Doug Ledford <dledford@redhat.com>

IB/core: Fix sparse warnings

Delete unused variables to prevent sparse warnings.

Fixes: db1b5ddd5336 ("IB/core: Rename uverbs event file structure")
Fixes: fd3c7904db6e ("IB/core: Change idr objects to use the new schema")

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

RDMA/bnxt_re: Fix the value reported for local ack delay

Local ack delay exposed by the driver is 0 which means infinite QP
timeout. Reporting the default value to 16 (approx 260ms)

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

RDMA/bnxt_re: Report MISSED_EVENTS in req_notify_cq

While invoking the req_notify_cq hook, ULPs can request
whether the CQs have any CQEs pending. If CQEs are pending,
drivers can indicate  it by returning 1 for req_notify_cq.
The stack will poll CQ again till CQ is empty.

This patch peeks the CQ for any valid entries and return accordingly.

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

RDMA/bnxt_re: Fix return value of poll routine

Fix the incorrect reporting of number of polled
entries by taking into account the max CQ depth
in the driver.

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

RDMA/bnxt_re: Enable atomics only if host bios supports

Driver shall check if the host system bios has enabled
Atomic operations capability in PCI Device Control 2
register of the pci-device. Expose the ATOMIC_HCA
flag only if the Atomic operations capability is set.

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

RDMA/bnxt_re: Specify RDMA component when allocating stats context

Starting FW version 20.6.47, firmware is keeping separate statistics
for L2 and RDMA. However, driver needs to specify RDMA or not when
allocating stat_ctx.

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

RDMA/bnxt_re: Fixed the max_rd_atomic support for initiator and destination QP

There's a couple of bugs in the support of max_rd_atomic and
max_dest_rd_atomic. In the modify_qp, if the requested max_rd_atomic,
which is the ORRQ size, is greater than what the chip can support,
then we have to cap the request to chip max as we can't have the HW
overflow the ORRQ. Capping the max_rd_atomic support internally is okay
to do as the remaining read/atomic WRs will still be sitting in the SQ.
However, for the max_dest_rd_atomic, the driver has to error out as
this dictates the IRRQ size and we can't control what the remote
side sends.

Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/bnxt_re: Report supported value to IB stack in query_device

- Report supported value for max_mr_size to IB stack in query_device.
   Also, check and log if MR size requested by application in
   reg_user_mr() is greater than value currently supported by driver.
 - Report only 4K page size support for now
 - Fix Max_QP value returned by ibv_devinfo -vv.
   In case of PF, FW reserves 129 QPs for creating QP1s of VFs
   and PF. So the max_qp value reported by FW for PF doesn'tt include
   the QP1. Fixing this issue by adding 1 with the value reported
   by FW.

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

RDMA/bnxt_re: Do not free the ctx_tbl entry if delete GID fails

This fix is added only to avoid system crash in some a
specific scenario. When bnxt_re driver is loaded and if
user tries to change interface mac address, delete GID
fails because QP1 is still associated with existing MAC
(default GID). If the above command fails GID tables are
not modified in the h/w or driver, but the GID context memory
is freed. Now, if the user changes the mac back to the original
value, another add_gid comes to the driver where the driver
reports that the GID is already present in its table
and tries to access the context which was already freed.

So, in this case, in order to  avoid NULL pointer de-reference,
this patch removes the context memory free  if delete_gid fails
and the same context memory is re-used in new add_gid.
Memory cleanup will be taken care during driver unload, while
deleting the GID table.

Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/bnxt_re: Fix WQE Size posted to HW to prevent it from throwing error

Posting WQE size of 2 results in a WQE_FORMAT_ERROR
thrown by the HW as it requires host to supply WQE Size with room
for atleast one SGE so that the resulting WQE size be atleast 3.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/bnxt_re: Free doorbell page index (DPI) during dealloc ucontext

The driver must free the DPI during the dealloc_ucontext
instead of freeing it during dealloc_pd. However, the DPI
allocation scheme remains unchanged.

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

IB/mlx5: Fix a warning message

"umem" is a valid pointer.  We intended to print "*umem" or even just
"err" instead.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/ocrdma: Fix error codes in ocrdma_create_srq()

If either of these allocations fail then we return ERR_PTR(0).  That's
equivalent to NULL and results in a NULL pointer dereference in the
caller.

Fixes: fe2caefcdf58 ("RDMA/ocrdma: Add driver for Emulex OneConnect IBoE RDMA adapter")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/ocrdma: Fix an error code in ocrdma_alloc_pd()

We should preserve the original "status" error code instead of resetting
it to zero.  Returning ERR_PTR(0) is the same as NULL and results in a
NULL dereference in the callers.  I added a printk() on error instead.

Fixes: 45e86b33ec8b ("RDMA/ocrdma: Cache recv DB until QP moved to RTR")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/cxgb3: Fix error codes in iwch_alloc_mr()

We accidentally don't set the error code on some error paths.  It means
return ERR_PTR(0) which is NULL and results in a NULL dereference in the
caller.

Fixes: 13a239330abd ("RDMA/cxgb3: Don't ignore insert_handle() failures")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

cxgb4: Fix error codes in c4iw_create_cq()

If one of these kmalloc() calls fails then we return ERR_PTR(0) which is
NULL.  It results in a NULL dereference in the callers.

Fixes: cfdda9d76436 ("RDMA/cxgb4: Add driver for Chelsio T4 RNIC")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/i40iw: Fix error code in i40iw_create_cq()

We accidentally forgot to set the error code if ib_copy_from_udata()
fails.  It means we return ERR_PTR(0) which is NULL and results in a
NULL dereference in the callers.

Fixes: d37498417947 ("i40iw: add files for iwarp interface")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/IPoIB: Fix error code in ipoib_add_port()

We accidentally don't see the error code on some of these error paths.
It means we return ERR_PTR(0) which is NULL and it results in a NULL
dereference in the caller.

This bug dates to pre-git days.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/bnxt_re: checking for NULL instead of IS_ERR()

bnxt_re_alloc_mw() doesn't return NULL, it returns error pointers.

Fixes: 9152e0b722b2 ("RDMA/bnxt_re: HW workarounds for handling specific conditions")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Free QP PBLEs when the QP is destroyed

If the physical buffer list entries (PBLEs) of a QP are freed
up at i40iw_dereg_mr, they can be assigned to a newly
created QP before the previous QP is destroyed. Fix this
by freeing PBLEs only when the QP is destroyed.

Signed-off-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Avoid memory leak of CQP request objects

Control Queue Pair (CQP) request objects, which have
not received a completion upon interface close, remain
in memory.

To fix this, identify and free all pending CQP request
objects during destroy CQP OP.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Update list correctly

To avoid infinite loop, in i40iw_ieq_handle_exception, update
plist inside while loop.

Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Add missing memory barrier

Add missing write memory barrier before writing the
header containing valid bit to the WQE in i40iw_puda_send.

Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Free QP resources on CQP destroy QP failure

Current flow leaves software QP structures in memory if
Control Queue Pair (CQP) destroy QP OP fails. To fix this,
free QP resources on fail of CQP destroy QP OP.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Release cm_id ref on PCI function reset

On PCI function reset, cm_id reference is not released
which causes an application hang, as it waits on the
cm_id to be released on rdma_destroy.

To fix this, call i40iw_cm_disconn during a PCI function
reset to clean-up resources and release cm_id reference.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Utilize iwdev->reset during PCI function reset

Utilize iwdev->reset on a PCI function reset notification
instead of passing in reset flag for resource clean-up.

Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Do not poll CCQ after it is destroyed

Control Queue Pair (CQP) OPs, in this case - Update SDs,
cannot poll the Control Completion Queue (CCQ) after CCQ is
destroyed. Instead, poll via registers.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

i40iw: Fix order of cleanup in close

The order for calling i40iw_destroy_pble_pool is incorrect.
Also, add PBLE_CHUNK_MEM init state to track pble pool
creation and destruction.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Henry Orosco <henry.orosco@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Allow QP state transition from reset to error

Playing with IP-O-IB interface can trigger a warning message:
"ib0: Failed to modify QP to ERROR state" to be logged.
This happens when the QP is in IB_QPS_RESET state and the stack
is trying to transition it to IB_QPS_ERR state in ipoib_ib_dev_stop().

According to the IB spec, Table 91 - "QP State Transition Properties"
it looks like the transition from reset to error is valid:

Transition: Any State to Error
Required Attributes: None
Optional Attributes: None allowed
Actions: Queue processing is stopped. Work Requests pending or in
process are completed in error, when possible.

This patch allows the transition and quiets the message.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hns: Fix for checkpatch.pl comment style warnings

This patch correct the comment style warnings caught by
checkpatch.pl script.

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

IB/hns: Fix the bug with modifying the MAC address without removing the driver

When modified the MAC address used hns_roce_mac function, we release and create
reserved qp again, It is not necessary to use spin_lock_bh and spin_unlock_bh in
handle_en_event, Otherwise, it will occur a error. This patch mainly fixes it.

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

IB/hns: Fix the bug with rdma operation

When opcode of work request is RDMA read and write, it
should use rdma_wr to get remote_addr and rkey. This
patch fixes it.

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

IB/hns: Fix the bug with wild pointer when destroy rc qp

When destroyed rc qp, the hr_qp will be used after freed. This patch
will fix it.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hns: Fix the bug of polling cq failed for loopback Qps

In hip06 SoC, RoCE driver creates 8 reserved loopback QPs to
ensure zero wqe when free mr. However, if the enabled phy
port number is less than 6, it will fail in polling cqe with
8 reserved loopback QPs.

In order to solve this problem, the number of loopback Qps
will be adjusted based on the number of enabled phy port.

Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Set dma_mask and coherent_dma_mask

The RXE coupled with dummy device causes to the kernel panic attached
below.  The panic happens when ib_register_device tries to set dma_mask
by accessing a NULLed parent device.

The RXE does not actually use DMA, so we can set the dma_mask
to architecture value.

[16240.199689] RIP: 0010:ib_register_device+0x468/0x5a0 [ib_core]
[16240.205289] RSP: 0018:ffffc9000220fc10 EFLAGS: 00010246
[16240.209909] RAX: 0000000000000024 RBX: ffff880220d1a2a8 RCX: 0000000000000000
[16240.212244] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000009
[16240.214385] RBP: ffffc9000220fcb0 R08: 0000000000000000 R09: 000000000000023f
[16240.254465] R10: 0000000000000007 R11: 0000000000000000 R12: 0000000000000000
[16240.259467] R13: 0000000000000000 R14: 0000000000000000 R15: ffff880220d1a2a8
[16240.263314] FS:  00007fd8ecca0740(0000) GS:ffff8802364c0000(0000) knlGS:0000000000000000
[16240.267292] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[16240.273503] CR2: 0000000000000218 CR3: 00000002253ba000 CR4: 00000000000006e0
[16240.277066] Call Trace:
[16240.281836]  ? __kmalloc+0x26f/0x280
[16240.286596]  rxe_register_device+0x297/0x300 [rdma_rxe]
[16240.291377]  rxe_add+0x535/0x5b0 [rdma_rxe]
[16240.297586]  rxe_net_add+0x3e/0xc0 [rdma_rxe]
[16240.302375]  rxe_param_set_add+0x65/0x144 [rdma_rxe]
[16240.307769]  param_attr_store+0x68/0xd0
[16240.311640]  module_attr_store+0x1d/0x30
[16240.316421]  sysfs_kf_write+0x3a/0x50
[16240.317802]  kernfs_fop_write+0xff/0x180
[16240.322989]  __vfs_write+0x37/0x140
[16240.328164]  ? handle_mm_fault+0xce/0x240
[16240.333340]  vfs_write+0xb2/0x1b0
[16240.335013]  SyS_write+0x55/0xc0
[16240.340632]  entry_SYSCALL_64_fastpath+0x1a/0xa9

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/rxe: Fix kernel panic from skb destructor

In the time between rxe_send has finished and skb destructor
called, the QP's ref count might be 0, leading to a possible
QP destruction. This will lead to a kernel panic when the destructor
dereferences the QP.

The operation of incrementing QP ref count at rxe_send and decrementing
from skb destructor will prevent this crash.

BUG: unable to handle kernel NULL pointer dereference at 000000000000072c
IP: [<ffffffffa05df765>] rxe_skb_tx_dtor+0x15/0x50 [rdma_rxe]
PGD 0 [16240.211178]
Oops: 0002 [#1] SMP
CPU: 3 PID: 0 Comm: swapper/3 Tainted: G           OE   4.9.0-mlnx #1
Hardware name: Red Hat KVM, BIOS Bochs 01/01/2011
task: ffff88042d6b1480 task.stack: ffffc90001904000
RIP: 0010:[<ffffffffa05df765>]  [<ffffffffa05df765>] rxe_skb_tx_dtor+0x15/0x50 [rdma_rxe]
RSP: 0018:ffff88043fcc3df0  EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff880429684700 RCX: ffff88042d248200
RDX: 00000000ffffffff RSI: 00000000fffffe01 RDI: ffff880429684700
RBP: ffff88043fcc3e00 R08: ffff88043fcda240 R09: 00000000ff2d1de6
R10: 0000000000000000 R11: 00000000f49cf6fe R12: ffff880429684700
R13: ffffffff81893f96 R14: ffffffff817d66f0 R15: ffff880427f74200
FS:  0000000000000000(0000) GS:ffff88043fcc0000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000000000072c CR3: 000000041d3df000 CR4: 00000000000006e0
Stack:
 ffffffff817b29cf ffff880429684700 ffff88043fcc3e18 ffffffff817b42c2
 ffff880429684700 ffff88043fcc3e40 ffffffff817b4332 ffff880429684700
 ffff880427f74238 ffff880427f74228 ffff88043fcc3e58 ffffffff81893f96
Call Trace:
 <IRQ> [16240.336345]  [<ffffffff817b29cf>] ? skb_release_head_state+0x4f/0xb0
 [<ffffffff817b42c2>] skb_release_all+0x12/0x30
 [<ffffffff817b4332>] kfree_skb+0x32/0x90
 [<ffffffff81893f96>] ndisc_error_report+0x36/0x40
 [<ffffffff817d4de1>] neigh_invalidate+0x81/0xf0
 [<ffffffff817d68f7>] neigh_timer_handler+0x207/0x2b0
 [<ffffffff81109295>] call_timer_fn+0x35/0x120
 [<ffffffff81109db7>] run_timer_softirq+0x1d7/0x460
 [<ffffffff8106155e>] ? kvm_sched_clock_read+0x1e/0x30
 [<ffffffff810366b9>] ? sched_clock+0x9/0x10
 [<ffffffff810cfed2>] ? sched_clock_cpu+0x72/0xa0
 [<ffffffff818dd537>] __do_softirq+0xd7/0x289
 [<ffffffff810a6c95>] irq_exit+0xb5/0xc0
 [<ffffffff818dd372>] smp_apic_timer_interrupt+0x42/0x50
 [<ffffffff818dc682>] apic_timer_interrupt+0x82/0x90
 <EOI> [16240.395776]  [<ffffffff818da156>] ? native_safe_halt+0x6/0x10
 [<ffffffff818d9e6e>] default_idle+0x1e/0xd0
 [<ffffffff8103797f>] arch_cpu_idle+0xf/0x20
 [<ffffffff818da2c5>] default_idle_call+0x35/0x40
 [<ffffffff810e3eb5>] cpu_startup_entry+0x185/0x210
 [<ffffffff81050433>] start_secondary+0x103/0x130
RIP  [<ffffffffa05df765>] rxe_skb_tx_dtor+0x15/0x50 [rdma_rxe]

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/ipoib: Let lower driver handle get_stats64 call

The driver checks if the lower level driver supports get_stats, and if
so calls it to get the updated statistics, otherwise takes from the
current netdevice stats object.

Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Add ordered workqueue for RoCE GID management

Currently the RoCE GID management uses the ib_wq to do add and delete new GIDs
according to the netdev events.

The ib_wq isn't an ordered workqueue and thus two work elements can be executed
concurrently which will result in unexpected behavior and inconsistency of the
GIDs cache content.

Example:
ifconfig eth1 11.11.11.11/16 up

This command will invoke the following netdev events in the following order:
1. NETDEV_UP
2. NETDEV_DOWN
3. NETDEV_UP

If (2) and (3) will be executed concurrently or in reverse order, instead of
having a new GID with 11.11.11.11 IP, we will end up without any new GIDs.

Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/mlx5: Clean mr_cache debugfs in case of failure

The failure in creation of debugfs entries for mr_cache left entries,
which were already created.

It caused to mismatch and misguiding for the end users. The solution
is to clean mr_cache debugfs root, so no leftovers will be in the
system. In addition, let's document why the error is not needed to be
forwarded to user in case of failure.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Remove NOIO QP create flag

There are no users for IB_QP_CREATE_USE_GFP_NOIO flag,
so let's remove it.

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

{net, IB}/mlx4: Remove gfp flags argument

The caller to the driver marks GFP_NOIO allocations with help
of memalloc_noio-* calls now. This makes redundant to pass down
to the driver gfp flags, which can be GFP_KERNEL only.

The patch removes the gfp flags argument and updates all driver paths.

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

IB/{rdmavt, qib, hfi1}: Remove gfp flags argument

The caller to the driver marks GFP_NOIO allocations with help
of memalloc_noio-* calls now. This makes redundant to pass down
to the driver gfp flags, which can be GFP_KERNEL only.

The patch removes the gfp flags argument and updates all driver paths.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/IPoIB: Convert IPoIB to memalloc_noio_* calls

Commit 21caf2fc1931 ("mm: teach mm by current context info to not do I/O
during memory allocation") added the memalloc_noio_(save|restore) functions
to enable people to modify the MM behavior by disabling I/O during memory
allocation. This was further extended in Fixes: 934f3072c17c ("mm: clear
__GFP_FS when PF_MEMALLOC_NOIO is set"). memalloc_noio_* functions prevent
allocation paths recursing back into the filesystem without explicitly
changing the flags for every allocation site.

However the IPoIB hasn't been keeping up with the changes and missed
completely these memalloc_noio_* calls. This led to update of
allocation site with special QP creation flag, see commit 09b93088d750
("IB: Add a QP creation flag to use GFP_NOIO allocations"), while this
flag is supported by small number of drivers in IB stack.

Let's change it by updating to memalloc_noio_* calls and allow
for every driver underneath enjoy NOIO allocations.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/IPoIB: Forward MTU change to driver below

This patch checks if there is a driver below that
needs to be updated on the new MTU and calls it
accordingly.

Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB: Convert msleep below 20ms to usleep_range

The msleep(1) may do not sleep 1 ms as expected
and will sleep longer. The simple conversion from
msleep to usleep_range between 1ms and 2ms can solve an
issue.

The full and comprehensive explanation can be found at [1] and [2].

[1] https://lkml.org/lkml/2007/8/3/250
[2] Documentation/timers/timers-howto.txt

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

IB/uverbs: Make use of ib_modify_qp variant to avoid resolving DMAC

This patch makes use of IB core's ib_modify_qp_with_udata function that
also resolves the DMAC and handles udata.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Introduce modify QP operation with udata

This patch adds new function ib_modify_qp_with_udata so that
uverbs layer can avoid handling L2 mac address at verbs layer
and depend on the core layer to resolve the mac address consistently
for all required QPs.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Don't resolve IP address to the loopback device

When resolving an IP address that is on the host of the caller the
result from querying the routing table is the loopback device. This is
not a valid response, because it doesn't represent the RDMA device and
the port.

Therefore, callers need to check the resolved device and if it is a
loopback device find an alternative way to resolve it. To avoid this we
make sure that the response from rdma_resolve_ip() will not be the
loopback device.

While that, we fix an static checker warning about dereferencing an
unintitialized pointer using the same solution as in commit abeffce90c7f
("net/mlx5e: Fix a -Wmaybe-uninitialized warning") as a reference.

Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/core: Namespace is mandatory input for address resolution

In function addr_resolve() the namespace is a required input parameter
and not an output. It is passed later for searching the routing table
and device addresses. Also, it shouldn't be copied back to the caller.

Fixes: 565edd1d5555 ('IB/addr: Pass network namespace as a parameter')
Cc: <stable@vger.kernel.org> # v4.3+
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/iser: Fix connection teardown race condition

Under heavy iser target(scst) start/stop stress during login/logout
on iser intitiator side happened trace call provided below.

The function iscsi_iser_slave_alloc iser_conn pointer could be NULL,
due to the fact that function iscsi_iser_conn_stop can be called before
and free iser connection. Let's protect that flow by introducing global mutex.

BUG: unable to handle kernel paging request at 0000000000001018
IP: [<ffffffffc0426f7e>] iscsi_iser_slave_alloc+0x1e/0x50 [ib_iser]
Call Trace:
? scsi_alloc_sdev+0x242/0x300
scsi_probe_and_add_lun+0x9e1/0xea0
? kfree_const+0x21/0x30
? kobject_set_name_vargs+0x76/0x90
? __pm_runtime_resume+0x5b/0x70
__scsi_scan_target+0xf6/0x250
scsi_scan_target+0xea/0x100
iscsi_user_scan_session.part.13+0x101/0x130 [scsi_transport_iscsi]
? iscsi_user_scan_session.part.13+0x130/0x130 [scsi_transport_iscsi]
iscsi_user_scan_session+0x1e/0x30 [scsi_transport_iscsi]
device_for_each_child+0x50/0x90
iscsi_user_scan+0x44/0x60 [scsi_transport_iscsi]
store_scan+0xa8/0x100
? common_file_perm+0x5d/0x1c0
dev_attr_store+0x18/0x30
sysfs_kf_write+0x37/0x40
kernfs_fop_write+0x12c/0x1c0
__vfs_write+0x18/0x40
vfs_write+0xb5/0x1a0
SyS_write+0x55/0xc0

Fixes: 318d311e8f01 ("iser: Accept arbitrary sg lists mapping if the device supports it")
Cc: <stable@vger.kernel.org> # v4.5+
Signed-off-by: Vladimir Neyelov <vladimirn@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimbeg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>

RDMA/core: Document confusing code

While looking into Coverity ID 1351047 I ran into the following
piece of code at
drivers/infiniband/core/verbs.c:496:

ret = rdma_addr_find_l2_eth_by_grh(&dgid, &sgid,
                                   ah_attr->dmac,
                                   wc->wc_flags & IB_WC_WITH_VLAN ?
                                   NULL : &vlan_id,
                                   &if_index, &hoplimit);

The issue here is that the position of arguments in the call to
rdma_addr_find_l2_eth_by_grh() function do not match the order of
the parameters:

&dgid is passed to sgid
&sgid is passed to dgid

This is the function prototype:

int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
   const union ib_gid *dgid,
   u8 *dmac, u16 *vlan_id, int *if_index,
   int *hoplimit)

My question here is if this is intentional?

Answer:
Yes. ib_init_ah_from_wc() creates ah from the incoming packet.
Incoming packet has dgid of the receiver node on which this code is
getting executed and sgid contains the GID of the sender.

When resolving mac address of destination, you use arrived dgid as
sgid and use sgid as dgid because sgid contains destinations GID whom to
respond to.

Signed-off-by: Gustavo A. R. Silva <garsilva@embeddedor.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

mlx5: Avoid that mlx5_ib_sg_to_klms() overflows the klms[] array

ib_map_mr_sg() can pass an SG-list to .map_mr_sg() that is larger
than what fits into a single MR. .map_mr_sg() must not attempt to
map more SG-list elements than what fits into a single MR.
Hence make sure that mlx5_ib_sg_to_klms() does not write outside
the MR klms[] array.

Fixes: b005d3164713 ("mlx5: Add arbitrary sg list support")
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Cc: Sagi Grimberg <sagi@grimberg.me>
Cc: Leon Romanovsky <leonro@mellanox.com>
Cc: Israel Rukshin <israelr@mellanox.com>
Cc: <stable@vger.kernel.org>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Doug Ledford <dledford@redhat.com>

IB/hfi1: Ensure dd->gi_mask can not be overflowed

As the code stands today the array access in remap_intr() is OK. To
future proof the code though we should explicitly check to ensure the
index value is not outside of the valid range. This is not a straight
forward calculation so err on the side of caution.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

Merge tag 'v4.13-rc1' into k.o/for-4.13-rc

Linux v4.13-rc1

Linux v4.13-rc1

Merge tag 'standardize-docs' of git://git.lwn.net/linux

Pull documentation format standardization from Jonathan Corbet:
 "This series converts a number of top-level documents to the RST format
  without incorporating them into the Sphinx tree. The hope is to bring
  some uniformity to kernel documentation and, perhaps more importantly,
  have our existing docs serve as an example of the desired formatting
  for those that will be added later.

  Mauro has gone through and fixed up a lot of top-level documentation
  files to make them conform to the RST format, but without moving or
  renaming them in any way. This will help when we incorporate the ones
  we want to keep into the Sphinx doctree, but the real purpose is to
  bring a bit of uniformity to our documentation and let the top-level
  docs serve as examples for those writing new ones"

* tag 'standardize-docs' of git://git.lwn.net/linux: (84 commits)
  docs: kprobes.txt: Fix whitespacing
  tee.txt: standardize document format
  cgroup-v2.txt: standardize document format
  dell_rbu.txt: standardize document format
  zorro.txt: standardize document format
  xz.txt: standardize document format
  xillybus.txt: standardize document format
  vfio.txt: standardize document format
  vfio-mediated-device.txt: standardize document format
  unaligned-memory-access.txt: standardize document format
  this_cpu_ops.txt: standardize document format
  svga.txt: standardize document format
  static-keys.txt: standardize document format
  smsc_ece1099.txt: standardize document format
  SM501.txt: standardize document format
  siphash.txt: standardize document format
  sgi-ioc4.txt: standardize document format
  SAK.txt: standardize document format
  rpmsg.txt: standardize document format
  robust-futexes.txt: standardize document format
  ...

Merge tag 'random_for_linus' of git://git./linux/kernel/git/tytso/random

Pull random updates from Ted Ts'o:
 "Add wait_for_random_bytes() and get_random_*_wait() functions so that
  callers can more safely get random bytes if they can block until the
  CRNG is initialized.

  Also print a warning if get_random_*() is called before the CRNG is
  initialized. By default, only one single-line warning will be printed
  per boot. If CONFIG_WARN_ALL_UNSEEDED_RANDOM is defined, then a
  warning will be printed for each function which tries to get random
  bytes before the CRNG is initialized. This can get spammy for certain
  architecture types, so it is not enabled by default"

* tag 'random_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/random:
  random: reorder READ_ONCE() in get_random_uXX
  random: suppress spammy warnings about unseeded randomness
  random: warn when kernel uses unseeded randomness
  net/route: use get_random_int for random counter
  net/neighbor: use get_random_u32 for 32-bit hash random
  rhashtable: use get_random_u32 for hash_rnd
  ceph: ensure RNG is seeded before using
  iscsi: ensure RNG is seeded before use
  cifs: use get_random_u32 for 32-bit lock random
  random: add get_random_{bytes,u32,u64,int,long,once}_wait family
  random: add wait_for_random_bytes() API

Merge branch 'work.mount' of git://git./linux/kernel/git/viro/vfs

Pull ->s_options removal from Al Viro:
 "Preparations for fsmount/fsopen stuff (coming next cycle). Everything
  gets moved to explicit ->show_options(), killing ->s_options off +
  some cosmetic bits around fs/namespace.c and friends. Basically, the
  stuff needed to work with fsmount series with minimum of conflicts
  with other work.

  It's not strictly required for this merge window, but it would reduce
  the PITA during the coming cycle, so it would be nice to have those
  bits and pieces out of the way"

* 'work.mount' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  isofs: Fix isofs_show_options()
  VFS: Kill off s_options and helpers
  orangefs: Implement show_options
  9p: Implement show_options
  isofs: Implement show_options
  afs: Implement show_options
  affs: Implement show_options
  befs: Implement show_options
  spufs: Implement show_options
  bpf: Implement show_options
  ramfs: Implement show_options
  pstore: Implement show_options
  omfs: Implement show_options
  hugetlbfs: Implement show_options
  VFS: Don't use save/replace_mount_options if not using generic_show_options
  VFS: Provide empty name qstr
  VFS: Make get_filesystem() return the affected filesystem
  VFS: Clean up whitespace in fs/namespace.c and fs/super.c
  Provide a function to create a NUL-terminated string from unterminated data

Merge branch 'work.__copy_to_user' of git://git./linux/kernel/git/viro/vfs

Pull more __copy_.._user elimination from Al Viro.

* 'work.__copy_to_user' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  drm_dp_aux_dev: switch to read_iter/write_iter

Merge branch 'work.uaccess-unaligned' of git://git./linux/kernel/git/viro/vfs

Pull uacess-unaligned removal from Al Viro:
 "That stuff had just one user, and an exotic one, at that - binfmt_flat
  on arm and m68k"

* 'work.uaccess-unaligned' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  kill {__,}{get,put}_user_unaligned()
  binfmt_flat: flat_{get,put}_addr_from_rp() should be able to fail

Merge branch 'misc.compat' of git://git./linux/kernel/git/viro/vfs

Pull network field-by-field copy-in updates from Al Viro:
 "This part of the misc compat queue was held back for review from
  networking folks and since davem has jus ACKed those..."

* 'misc.compat' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  get_compat_bpf_fprog(): don't copyin field-by-field
  get_compat_msghdr(): get rid of field-by-field copyin
  copy_msghdr_from_user(): get rid of field-by-field copyin