RDMA/core: Modify enum ib_gid_type and enum rdma_network_type

Separate IB_GID_TYPE_IB and IB_GID_TYPE_ROCE to two different values, so
enum ib_gid_type will match the gid types of the new query GID table API
which will be introduced in the following patches.

This change in enum ib_gid_type requires to change also enum
rdma_network_type by separating RDMA_NETWORK_IB and RDMA_NETWORK_ROCE_V1
values.

Link: https://lore.kernel.org/r/20200923165015.2491894-3-leon@kernel.org
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: Change rdma_get_gid_attr returned error code

Change the error code returned from rdma_get_gid_attr when the GID entry
is invalid but the GID index is in the gid table size range to -ENODATA
instead of -EINVAL.

This change is done in order to provide a more accurate error reporting to
be used by the new GID query API in user space. Nevertheless, -EINVAL is
still returned from sysfs in the aforementioned case to maintain
compatibility with user space that expects -EINVAL.

Link: https://lore.kernel.org/r/20200923165015.2491894-2-leon@kernel.org
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

overflow: Include header file with SIZE_MAX declaration

The various array_size functions use SIZE_MAX define, but missed limits.h
causes to failure to compile code that needs overflow.h.

 In file included from drivers/infiniband/core/uverbs_std_types_device.c:6:
 ./include/linux/overflow.h: In function 'array_size':
 ./include/linux/overflow.h:258:10: error: 'SIZE_MAX' undeclared (first use in this function)
   258 |   return SIZE_MAX;
       |          ^~~~~~~~

Fixes: 610b15c50e86 ("overflow.h: Add allocation size calculation helpers")
Link: https://lore.kernel.org/r/20200913102928.134985-1-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/qedr: Endianness warnings cleanup

Making a change to fix following sparse warnings reported by kbuild bot.

 CHECK   drivers/infiniband/hw/qedr/verbs.c

drivers/infiniband/hw/qedr/verbs.c:3872:59: warning: incorrect type in assignment (different base types)
drivers/infiniband/hw/qedr/verbs.c:3872:59:    expected restricted __le32 [usertype] sge_prod
drivers/infiniband/hw/qedr/verbs.c:3872:59:    got unsigned int [usertype] sge_prod
drivers/infiniband/hw/qedr/verbs.c:3875:59: warning: incorrect type in assignment (different base types)
drivers/infiniband/hw/qedr/verbs.c:3875:59:    expected restricted __le32 [usertype] wqe_prod
drivers/infiniband/hw/qedr/verbs.c:3875:59:    got unsigned int [usertype] wqe_prod

Link: https://lore.kernel.org/r/20201001100959.19940-1-palok@marvell.com
Reported-by: kbuild test robot <lkp@intel.com>
Fixes: acca72e2b031 ("RDMA/qedr: SRQ's bug fixes")
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Michal Kalderon <mkalderon@marvell.com>
Signed-off-by: Alok Prasad <palok@marvell.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Constify static struct attribute_group

The only usage of these is to pass their address to sysfs_create_group()
and sysfs_remove_group(), both which takes const pointers. Make it const
to allow the compiler to put them in read-only memory.

Link: https://lore.kernel.org/r/20200930224004.24279-3-rikard.falkeborn@gmail.com
Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Constify struct attribute_group

The only usage of the pma_table field in the ib_port struct is to pass its
address to sysfs_create_group() and sysfs_remove_group(). Make it const to
make it possible to constify a couple of static struct
attribute_group. This allows the compiler to put them in read-only memory.

Link: https://lore.kernel.org/r/20200930224004.24279-2-rikard.falkeborn@gmail.com
Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Sync device with CPU pages upon ODP MR registration

Sync device with CPU pages upon ODP MR registration. mlx5 already has to
zero the HW's version of the PAS list, may as well deliver a PAS list that
matches the current CPU page tables configuration.

Link: https://lore.kernel.org/r/20200930163828.1336747-5-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/mlx5: Extend advice MR to support non faulting mode

Extend advice MR to support non faulting mode, this can improve
performance by increasing the populated page tables in the device.

Link: https://lore.kernel.org/r/20200930163828.1336747-4-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>

IB/core: Enable ODP sync without faulting

Enable ODP sync without faulting, this improves performance by reducing
the number of page faults in the system.

The gain from this option is that the device page table can be aligned
with the presented pages in the CPU page table without causing page
faults.

As of that, the overhead on data path from hardware point of view to
trigger a fault which end-up by calling the driver to bring the pages
will be dropped.

Link: https://lore.kernel.org/r/20200930163828.1336747-3-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>

IB/core: Improve ODP to use hmm_range_fault()

Move to use hmm_range_fault() instead of get_user_pags_remote() to improve
performance in a few aspects:

This includes:
- Dropping the need to allocate and free memory to hold its output

- No need any more to use put_page() to unpin the pages

- The logic to detect contiguous pages is done based on the returned
  order, no need to run per page and evaluate.

In addition, moving to use hmm_range_fault() enables to reduce page faults
in the system with it's snapshot mode, this will be introduced in next
patches from this series.

As part of this, cleanup some flows and use the required data structures
to work with hmm_range_fault().

Link: https://lore.kernel.org/r/20200930163828.1336747-2-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/addr: Fix race with netevent_callback()/rdma_addr_cancel()

This three thread race can result in the work being run once the callback
becomes NULL:

       CPU1                 CPU2                   CPU3
 netevent_callback()
                     process_one_req()       rdma_addr_cancel()
                      [..]
     spin_lock_bh()
   set_timeout()
     spin_unlock_bh()

spin_lock_bh()
list_del_init(&req->list);
spin_unlock_bh()

     req->callback = NULL
     spin_lock_bh()
       if (!list_empty(&req->list))
                         // Skipped!
         // cancel_delayed_work(&req->work);
     spin_unlock_bh()

    process_one_req() // again
     req->callback() // BOOM
cancel_delayed_work_sync()

The solution is to always cancel the work once it is completed so any
in between set_timeout() does not result in it running again.

Cc: stable@vger.kernel.org
Fixes: 44e75052bc2a ("RDMA/rdma_cm: Make rdma_addr_cancel into a fence")
Link: https://lore.kernel.org/r/20200930072007.1009692-1-leon@kernel.org
Reported-by: Dan Aloni <dan@kernelim.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Remove ucontext->closing

Nothing reads this any more, and the reason for its existence has passed
due to the deferred fput() scheme.

Fixes: 8ea1f989aa07 ("drivers/IB,usnic: reduce scope of mmap_sem")
Link: https://lore.kernel.org/r/0-v1-df64ff042436+42-uctx_closing_jgg@nvidia.com
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Remove unused field of rtrs_iu

list field is not used anywhere

Link: https://lore.kernel.org/r/20200930131407.6438-1-gi-oh.kim@clous.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove unused variables and definitions

Some code was removed but the variables were still there, and some
parameters have been changed to be queried from firmware. So the
definitions of them are no longer needed.

Fixes: 2a3d923f8730 ("RDMA/hns: Replace magic numbers with #defines")
Fixes: 82e620d9c3a0 ("RDMA/hns: Modify the data structure of hns_roce_av")
Fixes: 82547469782a ("IB/hns: Implement the add_gid/del_gid and optimize the GIDs management")
Fixes: 21b97f538765 ("RDMA/hns: Fixup qp release bug")
Link: https://lore.kernel.org/r/1601371934-40003-1-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/i40iw: Remove intermediate pointer that points to the same struct

There is no real need to have an intermediate pointer for the same struct,
remove it, and use struct directly.

Link: https://lore.kernel.org/r/20200926102450.2966017-11-leon@kernel.org
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mthca: Combine special QP struct with mthca QP

As preparation for the removal of QP allocation logic, we need to ensure
that ib_core allocates the right amount of memory before a call to the
driver create_qp(). It requires from driver to have the same structs for
all types of QPs.

Link: https://lore.kernel.org/r/20200926102450.2966017-10-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/drivers: Remove udata check from special QP

GSI QP can't be created from the user space, hence the udata check is
always false (udata == NULL). Remove that check and simplify the flow.

Link: https://lore.kernel.org/r/20200926102450.2966017-9-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/core: Align write and ioctl checks of QP types

The ioctl flow checks that the user provides only a supported list of QP
types, while write flow didn't do it and relied on the driver to check
it. Align those flows to fail as early as possible.

Link: https://lore.kernel.org/r/20200926102450.2966017-8-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx4: Prepare QP allocation to remove from the driver

Since all mlx4 QP have same storage type, move the QP allocation to be in
one place. This change is preparation to removal of such allocation from
the driver.

Link: https://lore.kernel.org/r/20200926102450.2966017-7-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx4: Embed GSI QP into general mlx4_ib QP

Refactor the storage struct of mlx4 GSI QP to be embedded in mlx4_ib
QP. This allows to remove internal memory allocation of QP struct which is
hidden inside the mlx4_ib_create_qp() flow.

Link: https://lore.kernel.org/r/20200926102450.2966017-6-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Delete not needed GSI QP signal QP type

GSI QP doesn't need signal QP type because it is initialized statically to
zero, which is IB_SIGNAL_ALL_WR also wr->send_flags isn't set too.  This
means that the GSI QP signal QP type can be removed.

Link: https://lore.kernel.org/r/20200926102450.2966017-5-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Change GSI QP to have same creation flow like other QPs

There is no reason to have separate create flow for the GSI QP, while
general create_qp routine has all needed checks and ability to allocate
and free the proper struct mlx5_ib_qp.

Link: https://lore.kernel.org/r/20200926102450.2966017-4-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Reuse existing fields in parent QP storage object

Remove duplication of mlx5_ib_qp and mlx5_ib_gsi_qp fields.  This change
returns the memory footprint of mlx5_ib QP to be as it was before
embedding GSI QP.

Link: https://lore.kernel.org/r/20200926102450.2966017-3-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Embed GSI QP into general mlx5_ib QP

The GSI QPs have different create flow from the regular QPs, but it is not
really needed. Update the code to use mlx5_ib_qp as a storage class for
all outside of GSI calls.

Link: https://lore.kernel.org/r/20200926102450.2966017-2-leon@kernel.org
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Fix type warning of sizeof in __mlx5_ib_alloc_counters()

sizeof() when applied to a pointer typed expression should give the size
of the pointed data, even if the data is a pointer.

Fixes: e1f24a79f424 ("IB/mlx5: Support congestion related counters")
Link: https://lore.kernel.org/r/20200917081354.2083293-1-liushixin2@huawei.com
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Acked-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Support inline data in extented sge space for RC

HIP08 supports RC inline up to size of 32 Bytes, and all data should be
put into SQWQE. For HIP09, this capability is extended to 1024 Bytes, if
length of data is longer than 32 Bytes, they will be filled into extended
sge space.

Link: https://lore.kernel.org/r/1599744069-9968-1-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Fix missing sq_sig_type when querying QP

The sq_sig_type field should be filled when querying QP, or the users may
get a wrong value.

Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/1600509802-44382-9-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Fix configuration of ack_req_freq in QPC

The hardware will add AckReq flag in BTH header according to the value of
ack_req_freq to request ACK from responder for the packets with this flag.
It should be greater than or equal to lp_pktn_ini instead of using a fixed
value.

Fixes: 7b9bd73ed13d ("RDMA/hns: Fix wrong assignment of lp_pktn_ini in QPC")
Link: https://lore.kernel.org/r/1600509802-44382-8-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Fix the wrong value of rnr_retry when querying qp

The rnr_retry returned to the user is not correct, it should be got from
another fields in QPC.

Fixes: bfe860351e31 ("RDMA/hns: Fix cast from or to restricted __le32 for driver")
Link: https://lore.kernel.org/r/1600509802-44382-7-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: Solve the overflow of the calc_pg_sz()

calc_pg_sz() may gets a data calculation overflow if the PAGE_SIZE is 64 KB
and hop_num is 2. It is because that all variables involved in calculation
are defined in type of int. So change the type of bt_chunk_size,
buf_chunk_size and obj_per_chunk_default to u64.

Fixes: ba6bb7e97421 ("RDMA/hns: Add interfaces to get pf capabilities from firmware")
Link: https://lore.kernel.org/r/1600509802-44382-6-git-send-email-liweihang@huawei.com
Signed-off-by: Jiaran Zhang <zhangjiaran@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add check for the validity of sl configuration

According to the RoCE v1 specification, the sl (service level) 0-7 are
mapped directly to priorities 0-7 respectively, sl 8-15 are reserved. The
driver should verify whether the the value of sl is larger than 7, if so,
an exception should be returned.

Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/1600509802-44382-5-git-send-email-liweihang@huawei.com
Signed-off-by: Jiaran Zhang <zhangjiaran@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Correct typo of hns_roce_create_cq()

Change "initialze" to "initialize".

Fixes: 8f3e9f3ea087 ("IB/hns: Add code for refreshing CQ CI using TPTR")
Link: https://lore.kernel.org/r/1600509802-44382-4-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: Add interception for resizing SRQs

HIP08 doesn't support modifying the maximum number of outstanding WR in an
SRQ. However, the driver does not return a failure message, and users may
mistakenly think that the resizing is executed successfully. So the driver
needs to intercept this operation.

Fixes: ffb1308b88b6 ("RDMA/hns: Move SRQ code to the reasonable place")
Link: https://lore.kernel.org/r/1600509802-44382-3-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Refactor process about opcode in post_send()

According to the IB specifications, the verbs should return an immediate
error when the users set an unsupported opcode. Furthermore, refactor codes
about opcode in process of post_send to make the difference between opcodes
clearer.

Link: https://lore.kernel.org/r/1600509802-44382-2-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 for SCCC in size of 64 Bytes

For HIP09, size of SCCC (Soft Congestion Control Context) is increased to
64 Bytes from 32 Bytes. The hardware will get the configuration of SCCC
from driver instead of using a fixed value.

Link: https://lore.kernel.org/r/1600245806-56321-5-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add support for QPC in size of 512 Bytes

The new version of RoCEE supports using QPC in size of 256B or 512B, so
that HIP09 can supports new congestion control algorithms by using QPC in
larger size.

Link: https://lore.kernel.org/r/1600245806-56321-4-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add support for CQE in size of 64 Bytes

The new version of RoCEE supports using CQE in size of 32B or 64B. The
performance of bus can be improved by using larger size of CQE.

Link: https://lore.kernel.org/r/1600245806-56321-3-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: Add support for EQE in size of 64 Bytes

The new version of RoCEE supports using CEQE in size of 4B or 64B, AEQE in
size of 16B or 64B. The performance of bus can be improved by using larger
size of EQE.

Link: https://lore.kernel.org/r/1600245806-56321-2-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/efa: Drop double zeroing for sg_init_table()

sg_init_table() zeroes its first argument, so the allocation of that
argument doesn't have to.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
expression x,n,flags;
@@

x =
- kcalloc
+ kmalloc_array
  (n,sizeof(*x),flags)
...
sg_init_table(x,n)
// </smpl>

Link: https://lore.kernel.org/r/1600601186-7420-6-git-send-email-Julia.Lawall@inria.fr
Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Acked-by: Gal Pressman <galpress@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

i40iw: Add support to make destroy QP synchronous

Occasionally ib_write_bw crash is seen due to access of a pd object in
i40iw_sc_qp_destroy after it is freed. Destroy qp is not synchronous in
i40iw and thus the iwqp object could be referencing a pd object that is
freed by ib core as a result of successful return from i40iw_destroy_qp.

Wait in i40iw_destroy_qp till all QP references are released and destroy
the QP and its associated resources before returning.  Switch to use the
refcount API vs atomic API for lifetime management of the qp.

 RIP: 0010:i40iw_sc_qp_destroy+0x4b/0x120 [i40iw]
 [...]
 RSP: 0018:ffffb4a7042e3ba8 EFLAGS: 00010002
 RAX: 0000000000000000 RBX: 0000000000000001 RCX: dead000000000122
 RDX: ffffb4a7042e3bac RSI: ffff8b7ef9b1e940 RDI: ffff8b7efbf09080
 RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000
 R10: 8080808080808080 R11: 0000000000000010 R12: ffff8b7efbf08050
 R13: 0000000000000001 R14: ffff8b7f15042928 R15: ffff8b7ef9b1e940
 FS:  0000000000000000(0000) GS:ffff8b7f2fa00000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000400 CR3: 000000020d60a006 CR4: 00000000001606e0
 Call Trace:
  i40iw_exec_cqp_cmd+0x4d3/0x5c0 [i40iw]
  ? try_to_wake_up+0x1ea/0x5d0
  ? __switch_to_asm+0x40/0x70
  i40iw_process_cqp_cmd+0x95/0xa0 [i40iw]
  i40iw_handle_cqp_op+0x42/0x1a0 [i40iw]
  ? cm_event_handler+0x13c/0x1f0 [iw_cm]
  i40iw_rem_ref+0xa0/0xf0 [i40iw]
  cm_work_handler+0x99c/0xd10 [iw_cm]
  process_one_work+0x1a1/0x360
  worker_thread+0x30/0x380
  ? process_one_work+0x360/0x360
  kthread+0x10c/0x130
  ? kthread_park+0x80/0x80
  ret_from_fork+0x35/0x40

Fixes: d37498417947 ("i40iw: add files for iwarp interface")
Link: https://lore.kernel.org/r/20200916131811.2077-1-shiraz.saleem@intel.com
Reported-by: Kamal Heib <kheib@redhat.com>
Signed-off-by: Sindhu, Devale <sindhu.devale@intel.com>
Signed-off-by: Shiraz, Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/efa: Add messages and RDMA read work requests HW stats

Add separate stats types for send messages and RDMA read work requests.

Link: https://lore.kernel.org/r/20200915141449.8428-3-galpress@amazon.com
Signed-off-by: Daniel Kranzdorf <dkkranzd@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: Group keep alive received counter with other SW stats

The keep alive received counter is a software stat, keep it grouped with
all other software stats.  Since all stored stats are software stats,
remove the efa_sw_stats struct and use efa_stats instead.

Link: https://lore.kernel.org/r/20200915141449.8428-2-galpress@amazon.com
Reviewed-by: Daniel Kranzdorf <dkkranzd@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/restrack: Improve readability in task name management

Use rdma_restrack_set_name() and rdma_restrack_parent_name() instead of
tricky uses of rdma_restrack_attach_task()/rdma_restrack_uadd().

This uniformly makes all restracks add'd using rdma_restrack_add().

Link: https://lore.kernel.org/r/20200922091106.2152715-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/restrack: Simplify restrack tracking in kernel flows

Have a single rdma_restrack_add() that adds an entry, there is no reason
to split the user/kernel here, the rdma_restrack_set_task() is responsible
for this difference.

This patch prepares the code to the future requirement of making restrack
is mandatory for managing ib objects.

Link: https://lore.kernel.org/r/20200922091106.2152715-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/restrack: Count references to the verbs objects

Refactor the restrack code to make sure the kref inside the restrack entry
properly kref's the object in which it is embedded. This slight change is
needed for future conversions of MR and QP which are refcounted before the
release and kfree.

The ideal flow from ib_core perspective as follows:
* Allocate ib_* structure with rdma_zalloc_*.
* Set everything that is known to ib_core to that newly created object.
* Initialize kref with restrack help
* Call to driver specific allocation functions.
* Insert into restrack DB
....
* Return and release restrack with restrack_put.

Largely this means a rdma_restrack_new() should be called near allocating
the containing structure.

Link: https://lore.kernel.org/r/20200922091106.2152715-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Don't call to restrack recursively

The restrack is going to manage memory of all IB objects and must be
called before object is created. GSI QP in the mlx5_ib separated between
creating dummy interface and HW object beneath. This was achieved by
double call to ib_create_qp().

In order to skip such reentry call to internal driver create_qp code.

Link: https://lore.kernel.org/r/20200922091106.2152715-3-leon@kernel.org
Reviewed-by: Mark Zhang <markz@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Delete from restrack DB after successful destroy

Update the code to have similar destroy pattern like other IB objects.

This change create asymmetry to the rdma_id_private create flow to make
sure that memory is managed by restrack.

Link: https://lore.kernel.org/r/20200922091106.2152715-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/ucma: Rework ucma_migrate_id() to avoid races with destroy

ucma_destroy_id() assumes that all things accessing the ctx will do so via
the xarray. This assumption violated only in the case the FD is being
closed, then the ctx is reached via the ctx_list. Normally this is OK
since ucma_destroy_id() cannot run concurrenty with release(), however
with ucma_migrate_id() is involved this can violated as the close of the
2nd FD can run concurrently with destroy on the first:

                CPU0                      CPU1
        ucma_destroy_id(fda)
                                  ucma_migrate_id(fda -> fdb)
                                       ucma_get_ctx()
        xa_lock()
         _ucma_find_context()
         xa_erase()
        xa_unlock()
                                       xa_lock()
                                        ctx->file = new_file
                                        list_move()
                                       xa_unlock()
                                      ucma_put_ctx()

                                   ucma_close(fdb)
                                      _destroy_id()
                                      kfree(ctx)

        _destroy_id()
          wait_for_completion()
          // boom, ctx was freed

The ctx->file must be modified under the handler and xa_lock, and prior to
modification the ID must be rechecked that it is still reachable from
cur_file, ie there is no parallel destroy or migrate.

To make this work remove the double locking and streamline the control
flow. The double locking was obsoleted by the handler lock now directly
preventing new uevents from being created, and the ctx_list cannot be read
while holding fgets on both files. Removing the double locking also
removes the need to check for the same file.

Fixes: 88314e4dda1e ("RDMA/cma: add support for rdma_migrate_id()")
Link: https://lore.kernel.org/r/0-v1-05c5a4090305+3a872-ucma_syz_migrate_jgg@nvidia.com
Reported-and-tested-by: syzbot+cc6fc752b3819e082d0c@syzkaller.appspotmail.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Clarify what the UMR is for when creating MRs

Once a mkey is created it can be modified using UMR. This is desirable for
performance reasons. However, different hardware has restrictions on what
modifications are possible using UMR. Make sense of these checks:

- mlx5_ib_can_reconfig_with_umr() returns true if the access flags can be
  altered. Most cases create MRs using 0 access flags (now made clear by
  consistent use of set_mkc_access_pd_addr_fields()), but the old logic
  here was tormented. Make it clear that this is checking if the current
  access_flags can be modified using UMR to different access_flags. It is
  always OK to use UMR to change flags that all HW supports.

- mlx5_ib_can_load_pas_with_umr() returns true if UMR can be used to
  enable and update the PAS/XLT. Enabling requires updating the entity
  size, so UMR ends up completely disabled on this old hardware. Make it
  clear why it is disabled. FRWR, ODP and cache always requires
  mlx5_ib_can_load_pas_with_umr().

- mlx5_ib_pas_fits_in_mr() is used to tell if an existing MR can be
  resized to hold a new PAS list. This only works for cached MR's because
  we don't store the PAS list size in other cases.

To be very clear, arrange things so any pre-created MR's in the cache
check the newly requested access_flags before allowing the MR to leave the
cache. If UMR cannot set the required access_flags the cache fails to
create the MR.

This in turn means relaxed ordering and atomic are now correctly blocked
early for implicit ODP on older HW.

Link: https://lore.kernel.org/r/20200914112653.345244-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Disable IB_DEVICE_MEM_MGT_EXTENSIONS if IB_WR_REG_MR can't work

set_reg_wr() always fails if !umr_modify_entity_size_disabled because
mlx5_ib_can_use_umr() always fails. Without set_reg_wr() IB_WR_REG_MR
doesn't work and that means the device should not advertise
IB_DEVICE_MEM_MGT_EXTENSIONS.

Fixes: 841b07f99a47 ("IB/mlx5: Block MR WR if UMR is not possible")
Link: https://lore.kernel.org/r/20200914112653.345244-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Make mkeys always owned by the kernel's PD when not enabled

Any mkey that is not enabled and assigned to userspace should have the PD
set to a kernel owned PD.

When cache entries are created for the first time the PDN is set to 0,
which is probably a kernel PD, but be explicit.

When a MR is registered using the hybrid reg_create with UMR xlt & enable
the disabled mkey is pointing at the user PD, keep it pointing at the
kernel until a UMR enables it and sets the user PD.

Fixes: 9ec4483a3f0f ("IB/mlx5: Move MRs to a kernel PD when freeing them to the MR cache")
Link: https://lore.kernel.org/r/20200914112653.345244-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Use set_mkc_access_pd_addr_fields() in reg_create()

reg_create() open codes this helper, use the shared code.

Link: https://lore.kernel.org/r/20200914112653.345244-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Remove dead check for EAGAIN after alloc_mr_from_cache()

alloc_mr_from_cache() no longer returns EAGAIN, this is just dead code
now.

Fixes: aad719dcf379 ("RDMA/mlx5: Allow MRs to be created in the cache synchronously")
Link: https://lore.kernel.org/r/20200914112653.345244-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Set the unsupported wr opcode

hip06 does not support IB_WR_LOCAL_INV, so the ps_opcode should be set to
an invalid value instead of being left uninitialized.

Fixes: 9a4435375cd1 ("IB/hns: Add driver files for hns RoCE driver")
Fixes: a2f3d4479fe9 ("RDMA/hns: Avoid unncessary initialization")
Link: https://lore.kernel.org/r/1600350615-115217-1-git-send-email-oulijun@huawei.com
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/qedr: Fix resource leak in qedr_create_qp

When xa_insert() fails, the acquired resource in qedr_create_qp should
also be freed. However, current implementation does not handle the error.

Fix this by adding a new goto label that calls qedr_free_qp_resources.

Fixes: 1212767e23bb ("qedr: Add wrapping generic structure for qpidr and adjust idr routines.")
Link: https://lore.kernel.org/r/20200911125159.4577-1-keitasuzuki.park@sslab.ics.keio.ac.jp
Signed-off-by: Keita Suzuki <keitasuzuki.park@sslab.ics.keio.ac.jp>
Acked-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/iw_cxgb4: Disable delayed ack by default

Receive side delayed ack mode is needed only for certain area networks/
connections. Therefore disable it by default.

Link: https://lore.kernel.org/r/20200909153707.2795-1-bharat@chelsio.com
Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Merge branch 'mlx_sw_owner_v2' into rdma.git for-next

Leon Romanovsky says:

====================
This series from Alex extends software steering interface to support
devices with extra capability "sw_owner_2" which will replace existing
"sw_owner".
====================

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

* branch 'mlx5_sw_owner_v2:
  RDMA/mlx5: Expose TIR and QP ICM address for sw_owner_v2 devices
  RDMA/mlx5: Allow DM allocation for sw_owner_v2 enabled devices
  RDMA/mlx5: Add sw_owner_v2 bit capability

Merge branch 'mlx5_active_speed' into rdma.git for-next

Leon Romanovsky says:

====================
IBTA declares speed as 16 bits, but kernel stores it in u8. This series
fixes in-kernel declaration while keeping external interface intact.
====================

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

* branch 'mlx5_active_speed':
  RDMA: Fix link active_speed size
  RDMA/mlx5: Delete duplicated mlx5_ptys_width enum
  net/mlx5: Refactor query port speed functions

RDMA: Fix link active_speed size

According to the IB spec active_speed size should be u16 and not u8 as
before. Changing it to allow further extensions in offered speeds.

Link: https://lore.kernel.org/r/20200917090223.1018224-4-leon@kernel.org
Signed-off-by: Aharon Landau <aharonl@mellanox.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Expose TIR and QP ICM address for sw_owner_v2 devices

Expose the ICM address to access TIR and QP, this will allow sw_owned_v2
devices to steer traffic to TIRs and QPs same as done with sw_owner
capability.

Link: https://lore.kernel.org/r/20200903073857.1129166-4-leon@kernel.org
Signed-off-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Allow DM allocation for sw_owner_v2 enabled devices

sw_owner_v2 will replace sw_owner for future devices, this means that if
sw_owner_v2 is set sw_owner should be ignored and DM allocation is
required for sw_owner_v2 devices to function.

Link: https://lore.kernel.org/r/20200903073857.1129166-3-leon@kernel.org
Signed-off-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Add sw_owner_v2 bit capability

Added sw_owner_v2 which will be enabled for future devices,
replacing sw_owner bit.

Signed-off-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

RDMA: Convert RWQ table logic to ib_core allocation scheme

Move struct ib_rwq_ind_table allocation to ib_core.

Link: https://lore.kernel.org/r/20200902081623.746359-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA: Clean MW allocation and free flows

Move allocation and destruction of memory windows under ib_core
responsibility and clean drivers to ensure that no updates to MW
ib_core structures are done in driver layer.

Link: https://lore.kernel.org/r/20200902081623.746359-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mlx5: Delete duplicated mlx5_ptys_width enum

Combine two same enums to avoid duplication.

Signed-off-by: Aharon Landau <aharonl@mellanox.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

net/mlx5: Refactor query port speed functions

The functions mlx5_query_port_link_width_oper and
mlx5_query_port_ib_proto_oper are always called together, so combine them
to a new function called mlx5_query_port_oper to avoid duplication.

And while the mlx5i_get_port_settings is the same as
mlx5_query_port_oper therefore let's remove it.

According to the IB spec link_width_oper and ib_proto_oper should be u16
and not as written u8, so perform casting as a preparation to cross-RDMA
patch which will fix that type for all drivers in the RDMA subsystem.

Fixes: ada68c31ba9c ("net/mlx5: Introduce a new header file for physical port functions")
Signed-off-by: Aharon Landau <aharonl@mellanox.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

RDMA/cma: Fix use after free race in roce multicast join

The roce path triggers a work queue that continues to touch the id_priv
but doesn't hold any reference on it. Futher, unlike in the IB case, the
work queue is not fenced during rdma_destroy_id().

This can trigger a use after free if a destroy is triggered in the
incredibly narrow window after the queue_work and the work starting and
obtaining the handler_mutex.

The only purpose of this work queue is to run the ULP event callback from
the standard context, so switch the design to use the existing
cma_work_handler() scheme. This simplifies quite a lot of the flow:

- Use the cma_work_handler() callback to launch the work for roce. This
  requires generating the event synchronously inside the
  rdma_join_multicast(), which in turn means the dummy struct
  ib_sa_multicast can become a simple stack variable.

- cm_work_handler() used the id_priv kref, so we can entirely eliminate
  the kref inside struct cma_multicast. Since the cma_multicast never
  leaks into an unprotected work queue the kfree can be done at the same
  time as for IB.

- Eliminating the general multicast.ib requires using cma_set_mgid() in a
  few places to recompute the mgid.

Fixes: 3c86aa70bf67 ("RDMA/cm: Add RDMA CM support for IBoE devices")
Link: https://lore.kernel.org/r/20200902081122.745412-9-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Consolidate the destruction of a cma_multicast in one place

Two places were open coding this sequence, and also pull in
cma_leave_roce_mc_group() which was called only once.

Link: https://lore.kernel.org/r/20200902081122.745412-8-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Remove dead code for kernel rdmacm multicast

There is no kernel user of RDMA CM multicast so this code managing the
multicast subscription of the kernel-only internal QP is dead. Remove it.

This makes the bug fixes in the next patches much simpler.

Link: https://lore.kernel.org/r/20200902081122.745412-7-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Combine cma_ndev_work with cma_work

These are the same thing, except that cma_ndev_work doesn't have a state
transition. Signal no state transition by setting old_state and new_state
== 0.

In all cases the handler function should not be called once
rdma_destroy_id() has progressed passed setting the state.

Link: https://lore.kernel.org/r/20200902081122.745412-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Remove cma_comp()

The only place that still uses it is rdma_join_multicast() which is only
doing a sanity check that the caller hasn't done something wrong and
doesn't need the spinlock.

At least in the case of rdma_join_multicast() the information it needs
will remain until the ID is destroyed once it enters these
states. Similarly there is no reason to check for these specific states in
the handler callback, instead use the usual check for a destroyed id under
the handler_mutex.

Link: https://lore.kernel.org/r/20200902081122.745412-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Fix locking for the RDMA_CM_LISTEN state

There is a strange unlocked read of the ID state when checking for
reuseaddr. This is because an ID cannot be reusable once it becomes a
listening ID. Instead of using the state to exclude reuse, just clear it
as part of rdma_listen()'s flow to convert reusable into not reusable.

Once a ID goes to listen there is no way back out, and the only use of
reusable is on the bind_list check.

Finally, update the checks under handler_mutex to use READ_ONCE and audit
that once RDMA_CM_LISTEN is observed in a req callback it is stable under
the handler_mutex.

Link: https://lore.kernel.org/r/20200902081122.745412-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Make the locking for automatic state transition more clear

Re-organize things so the state variable is not read unlocked. The first
attempt to go directly from ADDR_BOUND immediately tells us if the ID is
already bound, if we can't do that then the attempt inside
rdma_bind_addr() to go from IDLE to ADDR_BOUND confirms the ID needs
binding.

Link: https://lore.kernel.org/r/20200902081122.745412-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/cma: Fix locking for the RDMA_CM_CONNECT state

It is currently a bit confusing, but the design is if the handler_mutex
is held, and the state is in RDMA_CM_CONNECT, then the state cannot leave
RDMA_CM_CONNECT without also serializing with the handler_mutex.

Make this clearer by adding a direct assertion, fixing the usage in
rdma_connect and generally using READ_ONCE to read the state value.

Link: https://lore.kernel.org/r/20200902081122.745412-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/ipoib: Convert to use DEFINE_SEQ_ATTRIBUTE macro

Use DEFINE_SEQ_ATTRIBUTE macro to simplify the code.

Link: https://lore.kernel.org/r/20200916025022.3992627-1-liushixin2@huawei.com
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/i40iw: Avoid typecast from void to pci_dev

hw_context stores a pci_device pointer. Use the correct type instead of
void * and avoid typecasts.

Link: https://lore.kernel.org/r/20200914123528.270382-1-parav@mellanox.com
Signed-off-by: Parav Pandit <parav@nvidia.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/qedr: Add support for user mode XRC-SRQ's

Implement the XRC specific verbs.  The additional QP type introduced new
logic to the rest of the verbs that now require distinguishing whether a
QP has an "RQ" or an "SQ" or both.

Link: https://lore.kernel.org/r/20200722102339.30104-1-ybason@marvell.com
Signed-off-by: Michal Kalderon <mkalderon@marvell.com>
Signed-off-by: Yuval Basson <ybason@marvell.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Linux 5.9-rc5

Merge tag 'armsoc-fixes' of git://git./linux/kernel/git/soc/soc

Pull ARM SoC fixes from Olof Johansson:
 "A collection of fixes I've been accruing over the last few weeks, none
  of them have been severe enough to warrant flushing the queue but it's
  been long enough now that it's a good idea to send them in.

  A handful of them are fixups for QSPI DT/bindings/compatibles, some
  smaller fixes for system DMA clock control and TMU interrupts on i.MX,
  a handful of fixes for OMAP, including a fix for DSI (display) on
  omap5"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (27 commits)
  arm64: dts: ns2: Fixed QSPI compatible string
  ARM: dts: BCM5301X: Fixed QSPI compatible string
  ARM: dts: NSP: Fixed QSPI compatible string
  ARM: dts: bcm: HR2: Fixed QSPI compatible string
  dt-bindings: spi: Fix spi-bcm-qspi compatible ordering
  ARM: dts: imx6sx: fix the pad QSPI1B_SCLK mux mode for uart3
  arm64: dts: imx8mp: correct sdma1 clk setting
  arm64: dts: imx8mq: Fix TMU interrupt property
  ARM: dts: imx7d-zii-rmu2: fix rgmii phy-mode for ksz9031 phy
  ARM: dts: vfxxx: Add syscon compatible with OCOTP
  ARM: dts: imx6q-logicpd: Fix broken PWM
  arm64: dts: imx: Add missing imx8mm-beacon-kit.dtb to build
  ARM: dts: imx6q-prtwd2: Remove unneeded i2c unit name
  ARM: dts: imx6qdl-gw51xx: Remove unneeded #address-cells/#size-cells
  ARM: dts: imx7ulp: Correct gpio ranges
  ARM: dts: ls1021a: fix QuadSPI-memory reg range
  arm64: defconfig: Enable ptn5150 extcon driver
  arm64: defconfig: Enable USB gadget with configfs
  ARM: configs: Update Integrator defconfig
  ARM: dts: omap5: Fix DSI base address and clocks
  ...

Merge tag 'usb-5.9-rc5' of git://git./linux/kernel/git/gregkh/usb

Pull USB/Thunderbolt fixes from Greg KH:
 "Here are some small USB and Thunderbolt driver fixes for 5.9-rc5.

  Nothing huge, just a number of bugfixes and new device ids for
  problems reported:

   - new USB serial driver ids

   - bug fixes for syzbot reported problems

   - typec driver fixes

   - thunderbolt driver fixes

   - revert of reported broken commit

  All of these have been in linux-next with no reported issues"

* tag 'usb-5.9-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  usb: typec: intel_pmc_mux: Do not configure SBU and HSL Orientation in Alternate modes
  usb: typec: intel_pmc_mux: Do not configure Altmode HPD High
  usb: core: fix slab-out-of-bounds Read in read_descriptors
  Revert "usb: dwc3: meson-g12a: fix shared reset control use"
  usb: typec: ucsi: acpi: Check the _DEP dependencies
  usb: typec: intel_pmc_mux: Un-register the USB role switch
  usb: Fix out of sync data toggle if a configured device is reconfigured
  USB: serial: option: support dynamic Quectel USB compositions
  USB: serial: option: add support for SIM7070/SIM7080/SIM7090 modules
  thunderbolt: Use maximum USB3 link rate when reclaiming if link is not up
  thunderbolt: Disable ports that are not implemented
  USB: serial: ftdi_sio: add IDs for Xsens Mti USB converter

Merge tag 'staging-5.9-rc5' of git://git./linux/kernel/git/gregkh/staging

Pull staging/IIO driver fixes from Greg KH:
 "Here are a number of staging and IIO driver fixes for 5.9-rc5.

  The majority of these are IIO driver fixes, to resolve a timestamp
  issue that was recently found to affect a bunch of IIO drivers.

  The other fixes in here are:

   - small IIO driver fixes

   - greybus driver fix

   - counter driver fix (came in through the IIO fixes tree)

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'staging-5.9-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging: (23 commits)
  iio: adc: mcp3422: fix locking on error path
  iio: adc: mcp3422: fix locking scope
  iio: adc: meson-saradc: Use the parent device to look up the calib data
  iio:adc:max1118 Fix alignment of timestamp and data leak issues
  iio:adc:ina2xx Fix timestamp alignment issue.
  iio:adc:ti-adc084s021 Fix alignment and data leak issues.
  iio:adc:ti-adc081c Fix alignment and data leak issues
  iio:magnetometer:ak8975 Fix alignment and data leak issues.
  iio:light:ltr501 Fix timestamp alignment issue.
  iio:light:max44000 Fix timestamp alignment and prevent data leak.
  iio:chemical:ccs811: Fix timestamp alignment and prevent data leak.
  iio:proximity:mb1232: Fix timestamp alignment and prevent data leak.
  iio:accel:mma7455: Fix timestamp alignment and prevent data leak.
  iio:accel:bmc150-accel: Fix timestamp alignment and prevent data leak.
  iio:accel:mma8452: Fix timestamp alignment and prevent data leak.
  iio: accel: kxsd9: Fix alignment of local buffer.
  iio: adc: rockchip_saradc: select IIO_TRIGGERED_BUFFER
  iio: adc: ti-ads1015: fix conversion when CONFIG_PM is not set
  counter: microchip-tcb-capture: check the correct variable
  iio: cros_ec: Set Gyroscope default frequency to 25Hz
  ...

Merge tag 'driver-core-5.9-rc5' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core fixes from Greg KH:
 "Here are some small driver core and debugfs fixes for 5.9-rc5

  Included in here are:

   - firmware loader memory leak fix

   - firmware loader testing fixes for non-EFI systems

   - device link locking fixes found by lockdep

   - kobject_del() bugfix that has been affecting some callers

   - debugfs minor fix

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'driver-core-5.9-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  test_firmware: Test platform fw loading on non-EFI systems
  PM: <linux/device.h>: fix @em_pd kernel-doc warning
  kobject: Drop unneeded conditional in __kobject_del()
  driver core: Fix device_pm_lock() locking for device links
  MAINTAINERS: Add the security document to SECURITY CONTACT
  driver code: print symbolic error code
  debugfs: Fix module state check condition
  kobject: Restore old behaviour of kobject_del(NULL)
  firmware_loader: fix memory leak for paged buffer

Merge tag 'arm-soc/for-5.9/devicetree-fixes' of https://github.com/Broadcom/stblinux into arm/fixes

This pull request contains Broadcom ARM-based SoCs Device Tree fixes for
5.9, please pull the following:

- Florian fixes the Broadcom QSPI controller binding such that the most
  specific compatible string is the left most one, and all existing
  in-tree users are updated as well.

* tag 'arm-soc/for-5.9/devicetree-fixes' of https://github.com/Broadcom/stblinux:
  arm64: dts: ns2: Fixed QSPI compatible string
  ARM: dts: BCM5301X: Fixed QSPI compatible string
  ARM: dts: NSP: Fixed QSPI compatible string
  ARM: dts: bcm: HR2: Fixed QSPI compatible string
  dt-bindings: spi: Fix spi-bcm-qspi compatible ordering

Link: https://lore.kernel.org/r/20200909211857.4144718-1-f.fainelli@gmail.com
Signed-off-by: Olof Johansson <olof@lixom.net>

Merge tag 'imx-fixes-5.9-2' of git://git./linux/kernel/git/shawnguo/linux into arm/fixes

i.MX fixes for 5.9, round 2:

- Fix the misspelling of 'interrupts' property in i.MX8MQ TMU DT node.
- Correct 'ahb' clock for i.MX8MP SDMA1 in device tree.
- Fix pad QSPI1B_SCLK mux mode for UART3 on i.MX6SX.

* tag 'imx-fixes-5.9-2' of git://git.kernel.org/pub/scm/linux/kernel/git/shawnguo/linux:
  ARM: dts: imx6sx: fix the pad QSPI1B_SCLK mux mode for uart3
  arm64: dts: imx8mp: correct sdma1 clk setting
  arm64: dts: imx8mq: Fix TMU interrupt property

Link: https://lore.kernel.org/r/20200909143844.GA25109@dragon
Signed-off-by: Olof Johansson <olof@lixom.net>

Merge tag 'omap-for-v5.9/fixes-rc3' of git://git./linux/kernel/git/tmlind/linux-omap into arm/fixes

Fixes for omaps for v5.9-rc cycle

Few fixes for omap based devices:

- Fix of_clk_get() error handling for omap-iommu

- Fix missing audio pinctrl entries for logicpd boards

- Fix video for logicpd-som-lv after switch to generic panels

- Fix omap5 DSI clocks base

* tag 'omap-for-v5.9/fixes-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap:
  ARM: dts: omap5: Fix DSI base address and clocks
  ARM: dts: logicpd-som-lv-baseboard: Fix missing video
  ARM: dts: logicpd-som-lv-baseboard: Fix broken audio
  ARM: dts: logicpd-torpedo-baseboard: Fix broken audio
  ARM: OMAP2+: Fix an IS_ERR() vs NULL check in _get_pwrdm()

Link: https://lore.kernel.org/r/pull-1599132064-54898@atomide.com
Signed-off-by: Olof Johansson <olof@lixom.net>

Merge tag 'char-misc-5.9-rc5' of git://git./linux/kernel/git/gregkh/char-misc

Pull char / misc driver fixes from Greg KH:
 "Here are a number of small driver fixes for 5.9-rc5

  Included in here are:

   - habanalabs driver fixes

   - interconnect driver fixes

   - soundwire driver fixes

   - dyndbg fixes for reported issues, and then reverts to fix it all up
     to a sane state.

   - phy driver fixes

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'char-misc-5.9-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  Revert "dyndbg: accept query terms like file=bar and module=foo"
  Revert "dyndbg: fix problem parsing format="foo bar""
  scripts/tags.sh: exclude tools directory from tags generation
  video: fbdev: fix OOB read in vga_8planes_imageblit()
  dyndbg: fix problem parsing format="foo bar"
  dyndbg: refine export, rename to dynamic_debug_exec_queries()
  dyndbg: give %3u width in pr-format, cosmetic only
  interconnect: qcom: Fix small BW votes being truncated to zero
  soundwire: fix double free of dangling pointer
  interconnect: Show bandwidth for disabled paths as zero in debugfs
  habanalabs: fix report of RAZWI initiator coordinates
  habanalabs: prevent user buff overflow
  phy: omap-usb2-phy: disable PHY charger detect
  phy: qcom-qmp: Use correct values for ipq8074 PCIe Gen2 PHY init
  soundwire: bus: fix typo in comment on INTSTAT registers
  phy: qualcomm: fix return value check in qcom_ipq806x_usb_phy_probe()
  phy: qualcomm: fix platform_no_drv_owner.cocci warnings

Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull kvm fixes from Paolo Bonzini:
 "A bit on the bigger side, mostly due to me being on vacation, then
  busy, then on parental leave, but there's nothing worrisome.

  ARM:
   - Multiple stolen time fixes, with a new capability to match x86
   - Fix for hugetlbfs mappings when PUD and PMD are the same level
   - Fix for hugetlbfs mappings when PTE mappings are enforced (dirty
     logging, for example)
   - Fix tracing output of 64bit values

  x86:
   - nSVM state restore fixes
   - Async page fault fixes
   - Lots of small fixes everywhere"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (25 commits)
  KVM: emulator: more strict rsm checks.
  KVM: nSVM: more strict SMM checks when returning to nested guest
  SVM: nSVM: setup nested msr permission bitmap on nested state load
  SVM: nSVM: correctly restore GIF on vmexit from nesting after migration
  x86/kvm: don't forget to ACK async PF IRQ
  x86/kvm: properly use DEFINE_IDTENTRY_SYSVEC() macro
  KVM: VMX: Don't freeze guest when event delivery causes an APIC-access exit
  KVM: SVM: avoid emulation with stale next_rip
  KVM: x86: always allow writing '0' to MSR_KVM_ASYNC_PF_EN
  KVM: SVM: Periodically schedule when unregistering regions on destroy
  KVM: MIPS: Change the definition of kvm type
  kvm x86/mmu: use KVM_REQ_MMU_SYNC to sync when needed
  KVM: nVMX: Fix the update value of nested load IA32_PERF_GLOBAL_CTRL control
  KVM: fix memory leak in kvm_io_bus_unregister_dev()
  KVM: Check the allocation of pv cpu mask
  KVM: nVMX: Update VMCS02 when L2 PAE PDPTE updates detected
  KVM: arm64: Update page shift if stage 2 block mapping not supported
  KVM: arm64: Fix address truncation in traces
  KVM: arm64: Do not try to map PUDs when they are folded into PMD
  arm64/x86: KVM: Introduce steal-time cap
  ...

Merge tag 'for-linus' of git://github.com/openrisc/linux

Pull OpenRISC fixes from Stafford Horne:
 "Fixes for compile issues pointed out by kbuild and one bug I found in
  initrd with the 5.9 patches"

* tag 'for-linus' of git://github.com/openrisc/linux:
  openrisc: Fix issue with get_user for 64-bit values
  openrisc: Fix cache API compile issue when not inlining
  openrisc: Reserve memblock for initrd

Merge tag 'seccomp-v5.9-rc5' of git://git./linux/kernel/git/kees/linux

Pull seccomp fixes from Kees Cook:
 "This fixes a rare race condition in seccomp when using TSYNC and
  USER_NOTIF together where a memory allocation would not get freed
  (found by syzkaller, fixed by Tycho).

  Additionally updates Tycho's MAINTAINERS and .mailmap entries for his
  new address"

* tag 'seccomp-v5.9-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
  seccomp: don't leave dangling ->notif if file allocation fails
  mailmap, MAINTAINERS: move to tycho.pizza
  seccomp: don't leak memory when filter install races

Merge tag 'libnvdimm-fix-v5.9-rc5' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull libnvdimm fix from Vishal Verma:
 "Fix detection of dax support for block devices.

  Previous fixes in this area, which only affected printing of debug
  messages, had an incorrect condition for detection of dax. This fix
  should finally do the right thing"

* tag 'libnvdimm-fix-v5.9-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  dax: fix detection of dax support for non-persistent memory block devices

Merge tag 'for-5.9-rc4-tag' of git://git./linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:
 "A few more fixes:

   - regression fix for a crash after failed snapshot creation

   - one more lockep fix: use nofs allocation when allocating missing
     device

   - fix reloc tree leak on degraded mount

   - make some extent buffer alignment checks less strict to mount
     filesystems created by btrfs-convert"

* tag 'for-5.9-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: fix NULL pointer dereference after failure to create snapshot
  btrfs: free data reloc tree on failed mount
  btrfs: require only sector size alignment for parent eb bytenr
  btrfs: fix lockdep splat in add_missing_dev

Merge tag '5.9-rc4-smb3-fix' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fix from Steve French:
 "A fix for lookup on DFS link when cifsacl or modefromsid is used"

* tag '5.9-rc4-smb3-fix' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: fix DFS mount with cifsacl/modefromsid

KVM: emulator: more strict rsm checks.

Don't ignore return values in rsm_load_state_64/32 to avoid
loading invalid state from SMM state area if it was tampered with
by the guest.

This is primarly intended to avoid letting guest set bits in EFER
(like EFER.SVME when nesting is disabled) by manipulating SMM save area.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20200827171145.374620-8-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

KVM: nSVM: more strict SMM checks when returning to nested guest

* check that guest is 64 bit guest, otherwise the SVM related fields
  in the smm state area are not defined

* If the SMM area indicates that SMM interrupted a running guest,
  check that EFER.SVME which is also saved in this area is set, otherwise
  the guest might have tampered with SMM save area, and so indicate
  emulation failure which should triple fault the guest.

* Check that that guest CPUID supports SVM (due to the same issue as above)

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20200827162720.278690-4-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

SVM: nSVM: setup nested msr permission bitmap on nested state load

This code was missing and was forcing the L2 run with L1's msr
permission bitmap

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20200827162720.278690-3-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

SVM: nSVM: correctly restore GIF on vmexit from nesting after migration

Currently code in svm_set_nested_state copies the current vmcb control
area to L1 control area (hsave->control), under assumption that
it mostly reflects the defaults that kvm choose, and later qemu
overrides  these defaults with L2 state using standard KVM interfaces,
like KVM_SET_REGS.

However nested GIF (which is AMD specific thing) is by default is true,
and it is copied to hsave area as such.

This alone is not a big deal since on VMexit, GIF is always set to false,
regardless of what it was on VM entry.  However in nested_svm_vmexit we
were first were setting GIF to false, but then we overwrite the control
fields with value from the hsave area.  (including the nested GIF field
itself if GIF virtualization is enabled).

Now on normal vm entry this is not a problem, since GIF is usually false
prior to normal vm entry, and this is the value that copied to hsave,
and then restored, but this is not always the case when the nested state
is loaded as explained above.

To fix this issue, move svm_set_gif after we restore the L1 control
state in nested_svm_vmexit, so that even with wrong GIF in the
saved L1 control area, we still clear GIF as the spec says.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20200827162720.278690-2-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

openrisc: Fix issue with get_user for 64-bit values

A build failure was raised by kbuild with the following error.

    drivers/android/binder.c: Assembler messages:
    drivers/android/binder.c:3861: Error: unrecognized keyword/register name `l.lwz ?ap,4(r24)'
    drivers/android/binder.c:3866: Error: unrecognized keyword/register name `l.addi ?ap,r0,0'

The issue is with 64-bit get_user() calls on openrisc.  I traced this to
a problem where in the internally in the get_user macros there is a cast
to long __gu_val this causes GCC to think the get_user call is 32-bit.
This binder code is really long and GCC allocates register r30, which
triggers the issue. The 64-bit get_user asm tries to get the 64-bit pair
register, which for r30 overflows the general register names and returns
the dummy register ?ap.

The fix here is to move the temporary variables into the asm macros.  We
use a 32-bit __gu_tmp for 32-bit and smaller macro and a 64-bit tmp in
the 64-bit macro.  The cast in the 64-bit macro has a trick of casting
through __typeof__((x)-(x)) which avoids the below warning.  This was
barrowed from riscv.

    arch/openrisc/include/asm/uaccess.h:240:8: warning: cast to pointer from integer of different size

I tested this in a small unit test to check reading between 64-bit and
32-bit pointers to 64-bit and 32-bit values in all combinations.  Also I
ran make C=1 to confirm no new sparse warnings came up.  It all looks
clean to me.

Link: https://lore.kernel.org/lkml/202008200453.ohnhqkjQ%25lkp@intel.com/
Signed-off-by: Stafford Horne <shorne@gmail.com>
Reviewed-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>

x86/kvm: don't forget to ACK async PF IRQ

Merge commit 26d05b368a5c0 ("Merge branch 'kvm-async-pf-int' into HEAD")
tried to adapt the new interrupt based async PF mechanism to the newly
introduced IDTENTRY magic but unfortunately it missed the fact that
DEFINE_IDTENTRY_SYSVEC() doesn't call ack_APIC_irq() on its own and
all DEFINE_IDTENTRY_SYSVEC() users have to call it manually.

As the result all multi-CPU KVM guest hang on boot when
KVM_FEATURE_ASYNC_PF_INT is present. The breakage went unnoticed because no
KVM userspace (e.g. QEMU) currently set it (and thus async PF mechanism
is currently disabled) but we're about to change that.

Fixes: 26d05b368a5c0 ("Merge branch 'kvm-async-pf-int' into HEAD")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200908135350.355053-3-vkuznets@redhat.com>
Tested-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

x86/kvm: properly use DEFINE_IDTENTRY_SYSVEC() macro

DEFINE_IDTENTRY_SYSVEC() already contains irqentry_enter()/
irqentry_exit().

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200908135350.355053-2-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

KVM: VMX: Don't freeze guest when event delivery causes an APIC-access exit

According to SDM 27.2.4, Event delivery causes an APIC-access VM exit.
Don't report internal error and freeze guest when event delivery causes
an APIC-access exit, it is handleable and the event will be re-injected
during the next vmentry.

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1597827327-25055-2-git-send-email-wanpengli@tencent.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>