platform/kernel/linux-rpi.git
3 years agoRDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for QP
Jason Gunthorpe [Sun, 15 Nov 2020 11:43:08 +0000 (13:43 +0200)]
RDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for QP

Delete custom logic in the QP in favor of more general variant.

Link: https://lore.kernel.org/r/20201115114311.136250-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/mlx5: Directly compute the PAS list for raw QP RQ's
Jason Gunthorpe [Sun, 15 Nov 2020 11:43:07 +0000 (13:43 +0200)]
RDMA/mlx5: Directly compute the PAS list for raw QP RQ's

The RQ WQ created when making a raw ethernet QP copies the PAS list from
a dummy QPC command created earlier in the flow. The WQC and QPC PAS lists
are not fully compatible as the page_offset is a different size.

Create the RQ WQ's PAS list directly and do not try to copy it from
another command structure.

Like the prior patch, this also means that badly aligned buffers were not
correctly rejected.

Link: https://lore.kernel.org/r/20201115114311.136250-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for WQ
Jason Gunthorpe [Sun, 15 Nov 2020 11:43:06 +0000 (13:43 +0200)]
RDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for WQ

This fixes a subtle bug, the WQ mailbox has only 5 bits to describe the
page_offset, while mlx5_ib_get_buf_offset() is hard wired to only work
with 6 bit page_offsets.

Thus it did not properly reject badly aligned buffers.

Fixes: 79b20a6c3014 ("IB/mlx5: Add receive Work Queue verbs")
Fixes: 0fb2ed66a14c ("IB/mlx5: Add create and destroy functionality for Raw Packet QP")
Link: https://lore.kernel.org/r/20201115114311.136250-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/mlx5: Use ib_umem_find_best_pgoff() for SRQ
Jason Gunthorpe [Sun, 15 Nov 2020 11:43:05 +0000 (13:43 +0200)]
RDMA/mlx5: Use ib_umem_find_best_pgoff() for SRQ

SRQ uses a quantized and scaled page_offset, which is another variation of
ib_umem_find_best_pgsz(). Add mlx5_umem_find_best_quantized_pgoff() to
perform this calculation for each mailbox. A macro shows how the
calculation is directly connected to the mailbox format.

This new routine replaces the limited mlx5_ib_cont_pages() and
mlx5_ib_get_buf_offset() pairing which would reject valid configurations
rather than adjust the page_size to make it work.

In turn this is much more aggressive about choosing large page sizes for
these objects and when THP is enabled it will now often find a single page
solution.

Link: https://lore.kernel.org/r/20201115114311.136250-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/efa: Remove .create_ah callback assignment
Gal Pressman [Sun, 15 Nov 2020 10:34:02 +0000 (12:34 +0200)]
RDMA/efa: Remove .create_ah callback assignment

Drivers now expose two callbacks for address handle creation, one for
uverbs and one for kverbs. EFA only supports uverbs so the .create_ah
assignment can be removed. Fix the core code caller to check the proper
function pointer.

Link: https://lore.kernel.org/r/20201115103404.48829-3-galpress@amazon.com
Signed-off-by: Gal Pressman <galpress@amazon.com>
Acked-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Add new PCI device ID matching for HIP09
Lang Cheng [Thu, 12 Nov 2020 13:19:44 +0000 (21:19 +0800)]
RDMA/hns: Add new PCI device ID matching for HIP09

The 200G device has a new device ID 0xA228, add it to the PCI table.

Link: https://lore.kernel.org/r/1605187184-26079-1-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>
3 years agoRDMA/cma: Add missing error handling of listen_id
Leon Romanovsky [Wed, 4 Nov 2020 14:40:07 +0000 (16:40 +0200)]
RDMA/cma: Add missing error handling of listen_id

Don't silently continue if rdma_listen() fails but destroy previously
created CM_ID and return an error to the caller.

Fixes: d02d1f5359e7 ("RDMA/cma: Fix deadlock destroying listen requests")
Link: https://lore.kernel.org/r/20201104144008.3808124-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/restrack: Store all special QPs in restrack DB
Leon Romanovsky [Wed, 4 Nov 2020 14:40:06 +0000 (16:40 +0200)]
RDMA/restrack: Store all special QPs in restrack DB

Special QPs (SMI and GSI) have different rules in regards of their QP
numbers. While all other QP numbers are unique per-device, the QP0 and QP1
are created per-port as requested by IBTA.

In multiple port devices, the number of SMI and GSI QPs with be equal to
the number ports.

$ rdma dev
0: ibp0s9: node_type ca fw 4.4.9999 node_guid 5254:00c0:fe12:3455 sys_image_guid 5254:00c0:fe12:3455
$ rdma link
0/1: ibp0s9/1: subnet_prefix fe80:0000:0000:0000 lid 13397 sm_lid 49151 lmc 0 state ACTIVE physical_state LINK_UP
0/2: ibp0s9/2: subnet_prefix fe80:0000:0000:0000 lid 13397 sm_lid 49151 lmc 0 state UNKNOWN physical_state UNKNOWN

Before:
$ rdma res show qp type SMI,GSI
link ibp0s9/1 lqpn 0 type SMI state RTS sq-psn 0 comm [ib_core]
link ibp0s9/1 lqpn 1 type GSI state RTS sq-psn 0 comm [ib_core]

After:
$ rdma res show qp type SMI,GSI
link ibp0s9/1 lqpn 0 type SMI state RTS sq-psn 0 comm [ib_core]
link ibp0s9/1 lqpn 1 type GSI state RTS sq-psn 0 comm [ib_core]
link ibp0s9/2 lqpn 0 type SMI state RTS sq-psn 0 comm [ib_core]
link ibp0s9/2 lqpn 1 type GSI state RTS sq-psn 0 comm [ib_core]

Link: https://lore.kernel.org/r/20201104144008.3808124-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/counter: Combine allocation and bind logic
Leon Romanovsky [Wed, 4 Nov 2020 14:40:05 +0000 (16:40 +0200)]
RDMA/counter: Combine allocation and bind logic

RDMA counters are allocated and bounded to QP immediately after that.
Only after this two step process they are really usable. By combining
the logic, we are ensuring that once counter is returned to the caller,
it will have everything set.

Link: https://lore.kernel.org/r/20201104144008.3808124-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Fix double free of the pointer to TSQ/TPQ
Weihang Li [Thu, 12 Nov 2020 11:29:42 +0000 (19:29 +0800)]
RDMA/hns: Fix double free of the pointer to TSQ/TPQ

A return statement is omitted after getting HEM table, then the newly
allocated pointer will be freed directly, which will cause a calltrace
when the driver was removed.

Fixes: d6d91e46210f ("RDMA/hns: Add support for configuring GMV table")
Link: https://lore.kernel.org/r/1605180582-46504-1-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agonvme-rdma: Use ibdev_to_node instead of dereferencing ->dma_device
Christoph Hellwig [Fri, 6 Nov 2020 18:19:35 +0000 (19:19 +0100)]
nvme-rdma: Use ibdev_to_node instead of dereferencing ->dma_device

->dma_device is a private implementation detail of the RDMA core.  Use the
ibdev_to_node helper to get the NUMA node for a ib_device instead of
poking into ->dma_device.

Link: https://lore.kernel.org/r/20201106181941.1878556-5-hch@lst.de
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA: Lift ibdev_to_node from rds to common code
Christoph Hellwig [Fri, 6 Nov 2020 18:19:34 +0000 (19:19 +0100)]
RDMA: Lift ibdev_to_node from rds to common code

Lift the ibdev_to_node from rds to common code and document it.

Link: https://lore.kernel.org/r/20201106181941.1878556-4-hch@lst.de
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/core: Remove ib_dma_{alloc,free}_coherent
Christoph Hellwig [Fri, 6 Nov 2020 18:19:37 +0000 (19:19 +0100)]
RDMA/core: Remove ib_dma_{alloc,free}_coherent

These two functions are entirely unused.

Link: https://lore.kernel.org/r/20201106181941.1878556-7-hch@lst.de
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/umem: Use ib_dma_max_seg_size instead of dma_get_max_seg_size
Christoph Hellwig [Fri, 6 Nov 2020 18:19:33 +0000 (19:19 +0100)]
RDMA/umem: Use ib_dma_max_seg_size instead of dma_get_max_seg_size

RDMA ULPs must not call DMA mapping APIs directly but instead use the
ib_dma_* wrappers.

Fixes: 0c16d9635e3a ("RDMA/umem: Move to allocate SG table from pages")
Link: https://lore.kernel.org/r/20201106181941.1878556-3-hch@lst.de
Reported-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMa/mthca: Work around -Wenum-conversion warning
Arnd Bergmann [Mon, 26 Oct 2020 21:12:30 +0000 (22:12 +0100)]
RDMa/mthca: Work around -Wenum-conversion warning

gcc points out a suspicious mixing of enum types in a function that
converts from MTHCA_OPCODE_* values to IB_WC_* values:

drivers/infiniband/hw/mthca/mthca_cq.c: In function 'mthca_poll_one':
drivers/infiniband/hw/mthca/mthca_cq.c:607:21: warning: implicit conversion from 'enum <anonymous>' to 'enum ib_wc_opcode' [-Wenum-conversion]
  607 |    entry->opcode    = MTHCA_OPCODE_INVALID;

Nothing seems to ever check for MTHCA_OPCODE_INVALID again, no idea if
this is meaningful, but it seems harmless as it deals with an invalid
input.

Remove MTHCA_OPCODE_INVALID and set the ib_wc_opcode to 0xFF, which is
still bogus, but at least doesn't make compiler warnings.

Fixes: 2a4443a69934 ("[PATCH] IB/mthca: fill in opcode field for send completions")
Link: https://lore.kernel.org/r/20201026211311.3887003-1-arnd@kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/core: Make FD destroy callback void
Leon Romanovsky [Wed, 4 Nov 2020 14:45:56 +0000 (16:45 +0200)]
RDMA/core: Make FD destroy callback void

All FD object destroy implementations return 0, so declare this callback
void.

Link: https://lore.kernel.org/r/20201104144556.3809085-3-leon@kernel.org
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/core: Postpone uobject cleanup on failure till FD close
Leon Romanovsky [Wed, 4 Nov 2020 14:45:55 +0000 (16:45 +0200)]
RDMA/core: Postpone uobject cleanup on failure till FD close

Remove the ib_is_destroyable_retryable() concept.

The idea here was to allow the drivers to forcibly clean the HW object
even if they otherwise didn't want to (eg because of usecnt). This was an
attempt to clean up in a world where drivers were not allowed to fail HW
object destruction.

Now that we are going back to allowing HW objects to fail destroy this
doesn't make sense. Instead if a uobject's HW object can't be destroyed it
is left on the uobject list and it is up to uverbs_destroy_ufile_hw() to
clean it. Multiple passes over the uobject list allow hidden dependencies
to be resolved. If that fails the HW driver is broken, throw a WARN_ON and
leak the HW object memory.

All the other tricky failure paths (eg on creation error unwind) have
already been updated to this new model.

Link: https://lore.kernel.org/r/20201104144556.3809085-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoIB/isert: Do not excplicitly check == false for bool
Zou Wei [Tue, 3 Nov 2020 11:57:54 +0000 (19:57 +0800)]
IB/isert: Do not excplicitly check == false for bool

It is not the kernel style, warning reported by coccicheck:

./ib_isert.c:1104:12-24: WARNING: Comparison to bool

Link: https://lore.kernel.org/r/1604404674-32998-1-git-send-email-zou_wei@huawei.com
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Zou Wei <zou_wei@huawei.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rxe: Remove VLAN code leftovers from RXE
Zhu Yanjun [Mon, 2 Nov 2020 14:13:42 +0000 (22:13 +0800)]
RDMA/rxe: Remove VLAN code leftovers from RXE

Since the commit fd49ddaf7e26 ("RDMA/rxe: prevent rxe creation on top of
vlan interface") does not permit rxe on top of vlan device, all the stuff
related with vlan should be removed.

Fixes: fd49ddaf7e26 ("RDMA/rxe: prevent rxe creation on top of vlan interface")
Link: https://lore.kernel.org/r/1604326422-18625-1-git-send-email-yanjunz@nvidia.com
Signed-off-by: Zhu Yanjun <yanjunz@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoIB/mlx5: Add support for NDR link speed
Meir Lichtinger [Mon, 26 Oct 2020 13:37:38 +0000 (15:37 +0200)]
IB/mlx5: Add support for NDR link speed

The IBTA specification has new speed - NDR. That speed supports signaling
rate of 100Gb. mlx5 IB driver translates link modes reported by ConnectX
device to IB speed and width. Added translation of new 100Gb, 200Gb and
400Gb link modes to NDR IB type and width of x1, x2 or x4 respectively.

Link: https://lore.kernel.org/r/20201026133738.1340432-3-leon@kernel.org
Signed-off-by: Meir Lichtinger <meirl@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoIB/core: Add support for NDR link speed
Meir Lichtinger [Mon, 26 Oct 2020 13:37:37 +0000 (15:37 +0200)]
IB/core: Add support for NDR link speed

Add new IBTA speed NDR, supporting signaling rate of 100Gb.

Link: https://lore.kernel.org/r/20201026133738.1340432-2-leon@kernel.org
Signed-off-by: Meir Lichtinger <meirl@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/ipoib: Add 50Gb and 100Gb link speeds to ethtool
Meir Lichtinger [Mon, 26 Oct 2020 13:29:04 +0000 (15:29 +0200)]
RDMA/ipoib: Add 50Gb and 100Gb link speeds to ethtool

The IBTA specification has new speeds - HDR and NDR, supporting signaling
rate of 50Gb and 100Gb respectively. ethtool support of ipoib driver
translates IB speed to signaling rate. Added translation of HDR and NDR IB
types to rates of 50Gb and 100Gb ethernet speed.

Link: https://lore.kernel.org/r/20201026132904.1338526-1-leon@kernel.org
Signed-off-by: Meir Lichtinger <meirl@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rxe,siw: Restore uverbs_cmd_mask IB_USER_VERBS_CMD_POST_SEND
Jason Gunthorpe [Fri, 30 Oct 2020 14:03:05 +0000 (11:03 -0300)]
RDMA/rxe,siw: Restore uverbs_cmd_mask IB_USER_VERBS_CMD_POST_SEND

These two drivers open code the call to POST_SEND and do not use the
rdma-core wrapper to do it, thus their usages was missed during the audit.

Both drivers use this as a doorbell to signal the kernel to start DMA.

Fixes: 628c02bf38aa ("RDMA: Remove uverbs cmds from drivers that don't use them")
Link: https://lore.kernel.org/r/0-v1-4608c5610afa+fb-uverbs_cmd_post_send_fix_jgg@nvidia.com
Reported-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/siw: Fix typo of EAGAIN not -EAGAIN in siw_cm_work_handler()
Zhang Qilong [Wed, 28 Oct 2020 12:25:09 +0000 (20:25 +0800)]
RDMA/siw: Fix typo of EAGAIN not -EAGAIN in siw_cm_work_handler()

The rv cannot be 'EAGAIN' in the previous path, we should use '-EAGAIN' to
check it. For example:

Call trace:
 ->siw_cm_work_handler
->siw_proc_mpareq
->siw_recv_mpa_rr

Link: https://lore.kernel.org/r/20201028122509.47074-1-zhangqilong3@huawei.com
Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com>
Reviewed-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Use ib_umem_find_best_pgsz() for mkc's
Jason Gunthorpe [Mon, 26 Oct 2020 13:23:14 +0000 (15:23 +0200)]
RDMA/mlx5: Use ib_umem_find_best_pgsz() for mkc's

Now that all the PAS arrays or UMR XLT's for mkcs are filled using
rdma_for_each_block() we can use the common ib_umem_find_best_pgsz()
algorithm.

Link: https://lore.kernel.org/r/20201026132314.1336717-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Split mlx5_ib_update_xlt() into ODP and non-ODP cases
Jason Gunthorpe [Mon, 26 Oct 2020 13:23:13 +0000 (15:23 +0200)]
RDMA/mlx5: Split mlx5_ib_update_xlt() into ODP and non-ODP cases

Mixing these together is just a mess, make a dedicated version,
mlx5_ib_update_mr_pas(), which directly loads the whole MTT for a non-ODP
MR.

The split out version can trivially use a simple loop with
rdma_for_each_block() which allows using the core code to compute the MR
pages and avoids seeking in the SGL list after each chunk as the
__mlx5_ib_populate_pas() call required.

Significantly speeds loading large MTTs.

Link: https://lore.kernel.org/r/20201026132314.1336717-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Split the WR setup out of mlx5_ib_update_xlt()
Jason Gunthorpe [Mon, 26 Oct 2020 13:23:12 +0000 (15:23 +0200)]
RDMA/mlx5: Split the WR setup out of mlx5_ib_update_xlt()

The memory allocation is quite complicated, and makes this function hard
to understand. Refactor things so that a function call sets up the WR, SG,
DMA mapping and buffer, further splitting that into buffer and DMA/wr.

This also slightly changes the buffer allocation logic to try an order 0
page allocation (with OOM warnings on) before going to the emergency page.

Link: https://lore.kernel.org/r/20201026132314.1336717-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Move xlt_emergency_page_mutex into mr.c
Jason Gunthorpe [Mon, 26 Oct 2020 13:23:11 +0000 (15:23 +0200)]
RDMA/mlx5: Move xlt_emergency_page_mutex into mr.c

This is the only user, so remove the wrappers.

Link: https://lore.kernel.org/r/20201026132314.1336717-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Change mlx5_ib_populate_pas() to use rdma_for_each_block()
Jason Gunthorpe [Mon, 26 Oct 2020 13:23:10 +0000 (15:23 +0200)]
RDMA/mlx5: Change mlx5_ib_populate_pas() to use rdma_for_each_block()

This routine converts the umem SGL into a list of fixed pages for DMA,
which is exactly what rdma_umem_for_each_dma_block() is for, use the
common code directly.

Link: https://lore.kernel.org/r/20201026132314.1336717-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Remove npages from mlx5_ib_cont_pages()
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:36 +0000 (15:19 +0200)]
RDMA/mlx5: Remove npages from mlx5_ib_cont_pages()

Most callers don't need this, and the few that do can get it as
ib_umem_num_pages(umem).

Link: https://lore.kernel.org/r/20201026131936.1335664-8-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Remove ncont from mlx5_ib_cont_pages()
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:35 +0000 (15:19 +0200)]
RDMA/mlx5: Remove ncont from mlx5_ib_cont_pages()

This is the same as ib_umem_num_dma_blocks(umem, 1UL << page_shift), have
the callers compute it directly.

Link: https://lore.kernel.org/r/20201026131936.1335664-7-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Remove order from mlx5_ib_cont_pages()
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:34 +0000 (15:19 +0200)]
RDMA/mlx5: Remove order from mlx5_ib_cont_pages()

Only alloc_mr_from_cache() needs order and can trivially compute it, so
lift it to the one call site and remove the NULL arguments.

Link: https://lore.kernel.org/r/20201026131936.1335664-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Move mlx5_ib_cont_pages() to the creation of the mlx5_ib_mr
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:33 +0000 (15:19 +0200)]
RDMA/mlx5: Move mlx5_ib_cont_pages() to the creation of the mlx5_ib_mr

For the user MR path, instead of calling this after getting the umem, call
it as part of creating the struct mlx5_ib_mr and distill its output to a
single page_shift stored inside the mr.

This avoids passing around the tuple of its output. Based on the umem and
page_shift, the output arguments can be computed using:

  count == ib_umem_num_pages(mr->umem)
  shift == mr->page_shift
  ncont == ib_umem_num_dma_blocks(mr->umem, 1 << mr->page_shift)
  order == order_base_2(ncont)

And since mr->page_shift == umem_odp->page_shift then ncont ==
ib_umem_num_dma_blocks() == ib_umem_odp_num_pages() for ODP umems.

Link: https://lore.kernel.org/r/20201026131936.1335664-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Remove mlx5_ib_mr->npages
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:32 +0000 (15:19 +0200)]
RDMA/mlx5: Remove mlx5_ib_mr->npages

This is the same value as ib_umem_num_pages(mr->umem), use that instead.

Link: https://lore.kernel.org/r/20201026131936.1335664-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Fix corruption of reg_pages in mlx5_ib_rereg_user_mr()
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:31 +0000 (15:19 +0200)]
RDMA/mlx5: Fix corruption of reg_pages in mlx5_ib_rereg_user_mr()

reg_pages should always contain mr->npage since when the mr is finally
de-reg'd it is always subtracted out.

If there were any error exits then mlx5_ib_rereg_user_mr() would leave the
reg_pages adjusted and this will cause it to be double subtracted
eventually.

The manipulation of reg_pages is inherently connected to the umem, so lift
it out of set_mr_fields() and only adjust it around creating/destroying a
umem.

reg_pages is only used for diagnostics in sysfs.

Fixes: 7d0cc6edcc70 ("IB/mlx5: Add MR cache for large UMR regions")
Link: https://lore.kernel.org/r/20201026131936.1335664-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/mlx5: Remove mlx5_ib_mr->order
Jason Gunthorpe [Mon, 26 Oct 2020 13:19:30 +0000 (15:19 +0200)]
RDMA/mlx5: Remove mlx5_ib_mr->order

The is only ever set to non-zero if the MR is from the cache, and if it is
cached then the order is in cached_ent->order.

Make it clearer that use_umr_mtt_update() only returns true for cached MRs
and remove the redundant data.

Link: https://lore.kernel.org/r/20201026131936.1335664-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Convert various random sprintf sysfs _show uses to sysfs_emit
Joe Perches [Thu, 8 Oct 2020 02:36:27 +0000 (19:36 -0700)]
RDMA: Convert various random sprintf sysfs _show uses to sysfs_emit

Manual changes for sysfs_emit as cocci scripts can't easily convert them.

Link: https://lore.kernel.org/r/ecde7791467cddb570c6f6d2c908ffbab9145cac.1602122880.git.joe@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Manual changes for sysfs_emit and neatening
Joe Perches [Thu, 8 Oct 2020 02:36:26 +0000 (19:36 -0700)]
RDMA: Manual changes for sysfs_emit and neatening

Make changes to use sysfs_emit in the RDMA code as cocci scripts can not
be written to handle _all_ the possible variants of various sprintf family
uses in sysfs show functions.

While there, make the code more legible and update its style to be more
like the typical kernel styles.

Miscellanea:

o Use intermediate pointers for dereferences
o Add and use string lookup functions
o return early when any intermediate call fails so normal return is
  at the bottom of the function
o mlx4/mcg.c:sysfs_show_group: use scnprintf to format intermediate strings

Link: https://lore.kernel.org/r/f5c9e4c9d8dafca1b7b70bd597ee7f8f219c31c8.1602122880.git.joe@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/hns: Add support for filling GMV table
Weihang Li [Sat, 24 Oct 2020 03:07:16 +0000 (11:07 +0800)]
RDMA/hns: Add support for filling GMV table

Add a interface to fill GMV(SGID/SMAC/VLAN) table for HIP09, all of above
source address information is stored as an entry in GMV table. The users
just need to provide the index to the hardware when POST SEND.

Link: https://lore.kernel.org/r/1603508836-33054-3-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/hns: Add support for configuring GMV table
Weihang Li [Sat, 24 Oct 2020 03:07:15 +0000 (11:07 +0800)]
RDMA/hns: Add support for configuring GMV table

HIP09 supports to store SGID/SMAC/VLAN together in a table named GMV. The
driver needs to allocate memory for it and tell the information about this
region to hardware.

Link: https://lore.kernel.org/r/1603508836-33054-2-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Remove 'addr' from rtrs_clt_add_path_to_arr
Guoqing Jiang [Fri, 23 Oct 2020 07:43:53 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Remove 'addr' from rtrs_clt_add_path_to_arr

Remove the argument since it is not used in the function.

Link: https://lore.kernel.org/r/20201023074353.21946-13-jinpu.wang@cloud.ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs: Introduce rtrs_post_send
Guoqing Jiang [Fri, 23 Oct 2020 07:43:52 +0000 (09:43 +0200)]
RDMA/rtrs: Introduce rtrs_post_send

Since the three functions share the similar logic, let's introduce one
common function for it.

Link: https://lore.kernel.org/r/20201023074353.21946-12-jinpu.wang@cloud.ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-srv: Kill rtrs_srv_change_state_get_old
Guoqing Jiang [Fri, 23 Oct 2020 07:43:51 +0000 (09:43 +0200)]
RDMA/rtrs-srv: Kill rtrs_srv_change_state_get_old

This function isn't needed since no caller checks the old_state of sess.

Link: https://lore.kernel.org/r/20201023074353.21946-11-jinpu.wang@cloud.ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Remove duplicated code
Gioh Kim [Fri, 23 Oct 2020 07:43:50 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Remove duplicated code

process_info_rsp checks that sg_cnt is zero twice.

Link: https://lore.kernel.org/r/20201023074353.21946-10-jinpu.wang@cloud.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Remove duplicated switch-case handling for CM error events
Gioh Kim [Fri, 23 Oct 2020 07:43:49 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Remove duplicated switch-case handling for CM error events

The events returning the same error value are put together.

Link: https://lore.kernel.org/r/20201023074353.21946-9-jinpu.wang@cloud.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs: Remove unnecessary argument dir of rtrs_iu_free
Gioh Kim [Fri, 23 Oct 2020 07:43:48 +0000 (09:43 +0200)]
RDMA/rtrs: Remove unnecessary argument dir of rtrs_iu_free

The direction of DMA operation is already in the rtrs_iu

Link: https://lore.kernel.org/r/20201023074353.21946-8-jinpu.wang@cloud.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-srv: Fix typo
Guoqing Jiang [Fri, 23 Oct 2020 07:43:47 +0000 (09:43 +0200)]
RDMA/rtrs-srv: Fix typo

It should mean region here.

Fixes: 9cb837480424 ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20201023074353.21946-7-jinpu.wang@cloud.ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-srv: Don't guard the whole __alloc_srv with srv_mutex
Guoqing Jiang [Fri, 23 Oct 2020 07:43:46 +0000 (09:43 +0200)]
RDMA/rtrs-srv: Don't guard the whole __alloc_srv with srv_mutex

The purpose of srv_mutex is to protect srv_list as in put_srv, so no need
to hold it when allocate memory for srv since it could be time consuming.

Otherwise if one machine has limited memory, rsrv_close_work could be
blocked for a longer time due to the mutex is held by get_or_create_srv
since it can't get memory in time.

  INFO: task kworker/1:1:27478 blocked for more than 120 seconds.
        Tainted: G           O    4.14.171-1-storage #4.14.171-1.3~deb9
  "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
  kworker/1:1     D    0 27478      2 0x80000000
  Workqueue: rtrs_server_wq rtrs_srv_close_work [rtrs_server]
  Call Trace:
   ? __schedule+0x38c/0x7e0
   schedule+0x32/0x80
   schedule_preempt_disabled+0xa/0x10
   __mutex_lock.isra.2+0x25e/0x4d0
   ? put_srv+0x44/0x100 [rtrs_server]
   put_srv+0x44/0x100 [rtrs_server]
   rtrs_srv_close_work+0x16c/0x280 [rtrs_server]
   process_one_work+0x1c5/0x3c0
   worker_thread+0x47/0x3e0
   kthread+0xfc/0x130
   ? trace_event_raw_event_workqueue_execute_start+0xa0/0xa0
   ? kthread_create_on_node+0x70/0x70
   ret_from_fork+0x1f/0x30

Let's move all the logics from __find_srv_and_get and __alloc_srv to
get_or_create_srv, and remove the two functions. Then it should be safe
for multiple processes to access the same srv since it is protected with
srv_mutex.

And since we don't want to allocate chunks with srv_mutex held, let's
check the srv->refcount after get srv because the chunks could not be
allocated yet.

Fixes: 9cb837480424 ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20201023074353.21946-6-jinpu.wang@cloud.ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Missing error from rtrs_rdma_conn_established
Gioh Kim [Fri, 23 Oct 2020 07:43:45 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Missing error from rtrs_rdma_conn_established

When rtrs_rdma_conn_established returns error (non-zero value), the error
value is stored in con->cm_err and it cannot trigger
rtrs_rdma_error_recovery. Finally the error of rtrs_rdma_con_established
will be forgot.

Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/20201023074353.21946-5-jinpu.wang@cloud.ionos.com
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Avoid run destroy_con_cq_qp/create_con_cq_qp in parallel
Jack Wang [Fri, 23 Oct 2020 07:43:44 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Avoid run destroy_con_cq_qp/create_con_cq_qp in parallel

It could happen two kworkers race with each other:

        CPU0                             CPU1
    addr_resolver kworker           reconnect kworker
    rtrs_clt_rdma_cm_handler
    rtrs_rdma_addr_resolved
    create_con_cq_qp: s.dev_ref++
    "s.dev_ref is 1"
                                    wait in create_cm fails with TIMEOUT
                                    destroy_con_cq_qp: --s.dev_ref
                                    "s.dev_ref is 0"
                                    destroy_con_cq_qp: sess->s.dev = NULL
     rtrs_cq_qp_create -> create_qp(con, sess->dev->ib_pd...)
    sess->dev is NULL, panic.

To fix the problem using mutex to serialize create_con_cq_qp and
destroy_con_cq_qp.

Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/20201023074353.21946-4-jinpu.wang@cloud.ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Remove outdated comment in create_con_cq_qp
Jack Wang [Fri, 23 Oct 2020 07:43:43 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Remove outdated comment in create_con_cq_qp

As run destroy_con_cq_qp many times doesn't work, remove the comments.

Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/20201023074353.21946-3-jinpu.wang@cloud.ionos.com
Suggested-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rtrs-clt: Remove destroy_con_cq_qp in case route resolving failed
Danil Kipnis [Fri, 23 Oct 2020 07:43:42 +0000 (09:43 +0200)]
RDMA/rtrs-clt: Remove destroy_con_cq_qp in case route resolving failed

We call destroy_con_cq_qp(con) in rtrs_rdma_addr_resolved() in case route
couldn't be resolved and then again in create_cm() because nothing
happens.

Don't call destroy_con_cq_qp from rtrs_rdma_addr_resolved, create_cm()
does the clean up already.

Fixes: 6a98d71daea1 ("RDMA/rtrs: client: main functionality")
Link: https://lore.kernel.org/r/20201023074353.21946-2-jinpu.wang@cloud.ionos.com
Signed-off-by: Danil Kipnis <danil.kipnis@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/hns: Support owner mode doorbell
Lang Cheng [Tue, 20 Oct 2020 12:04:53 +0000 (20:04 +0800)]
RDMA/hns: Support owner mode doorbell

The doorbell needs to store PI information into QPC, so the RoCEE should
wait for the results of storing, that is, it needs two bus operations to
complete a doorbell. When ROCEE is in SDI mode, multiple doorbells may be
interlocked because the RoCEE can only handle bus operations serially. So a
flag to mark if HIP09 is working in SDI mode is added. When the SDI flag is
set, the ROCEE will ignore the PI information of the doorbell, continue to
fetch wqe and verify its validity by it's owner_bit.

Link: https://lore.kernel.org/r/1603195493-22741-1-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>
4 years agoIB/isert: add module param to set sg_tablesize for IO cmd
Max Gurtovoy [Mon, 19 Oct 2020 09:46:28 +0000 (12:46 +0300)]
IB/isert: add module param to set sg_tablesize for IO cmd

Currently, iser target support max IO size of 16MiB by default. For some
adapters, allocating this amount of resources might reduce the total
number of possible connections that can be created. For those adapters,
it's preferred to reduce the max IO size to be able to create more
connections. Since there is no handshake procedure for max IO size in iser
protocol, set the default max IO size to 1MiB and add a module parameter
for enabling the option to control it for suitable adapters.

Fixes: 317000b926b0 ("IB/isert: allocate RW ctxs according to max IO size")
Link: https://lore.kernel.org/r/20201019094628.17202-1-mgurtovoy@nvidia.com
Reported-by: Krishnamraju Eraparaju <krishna2@chelsio.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rxe: Compute PSN windows correctly
Bob Pearson [Tue, 13 Oct 2020 17:07:42 +0000 (12:07 -0500)]
RDMA/rxe: Compute PSN windows correctly

The code which limited the number of unacknowledged PSNs was incorrect.
The PSNs are limited to 24 bits and wrap back to zero from 0x00ffffff.
The test was computing a 32 bit value which wraps at 32 bits so that
qp->req.psn can appear smaller than the limit when it is actually larger.

Replace '>' test with psn_compare which is used for other PSN comparisons
and correctly handles the 24 bit size.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Link: https://lore.kernel.org/r/20201013170741.3590-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/core: Fix error return in _ib_modify_qp()
Jing Xiangfeng [Fri, 16 Oct 2020 07:58:45 +0000 (15:58 +0800)]
RDMA/core: Fix error return in _ib_modify_qp()

Fix to return error code PTR_ERR() from the error handling case instead of
0.

Fixes: 51aab12631dd ("RDMA/core: Get xmit slave for LAG")
Link: https://lore.kernel.org/r/20201016075845.129562-1-jingxiangfeng@huawei.com
Signed-off-by: Jing Xiangfeng <jingxiangfeng@huawei.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoIB/verbs: avoid nested container_of()
Arnd Bergmann [Mon, 26 Oct 2020 16:15:39 +0000 (17:15 +0100)]
IB/verbs: avoid nested container_of()

Nested container_of() calls work correctly but cause a warning when
building with W=2. Invoking it from an inline function like in
drivers/infiniband/hw/mlx5/mlx5_ib.h means we get hundreds of warnings
like:

include/linux/kernel.h:852:8: warning: declaration of '__mptr' shadows a previous local [-Wshadow]
  852 |  void *__mptr = (void *)(ptr);     \
      |        ^~~~~~
include/rdma/uverbs_ioctl.h:651:11: note: in expansion of macro 'container_of'
  651 |  (udata ? container_of(container_of(udata, struct uverbs_attr_bundle,   \
      |           ^~~~~~~~~~~~
include/rdma/uverbs_ioctl.h:651:24: note: in expansion of macro 'container_of'
  651 |  (udata ? container_of(container_of(udata, struct uverbs_attr_bundle,   \
      |                        ^~~~~~~~~~~~
drivers/infiniband/hw/mthca/mthca_qp.c:564:35: note: in expansion of macro 'rdma_udata_to_drv_context'
  564 |  struct mthca_ucontext *context = rdma_udata_to_drv_context(
      |                                   ^~~~~~~~~~~~~~~~~~~~~~~~~
include/linux/kernel.h:852:8: note: shadowed declaration is here
  852 |  void *__mptr = (void *)(ptr);     \
      |        ^~~~~~
include/rdma/uverbs_ioctl.h:651:11: note: in expansion of macro 'container_of'
  651 |  (udata ? container_of(container_of(udata, struct uverbs_attr_bundle,   \
      |           ^~~~~~~~~~~~
drivers/infiniband/hw/mthca/mthca_qp.c:564:35: note: in expansion of macro 'rdma_udata_to_drv_context'
  564 |  struct mthca_ucontext *context = rdma_udata_to_drv_context(
      |                                   ^~~~~~~~~~~~~~~~~~~~~~~~~
In file included from <command-line>:
include/linux/kernel.h:852:8: warning: declaration of '__mptr' shadows a previous local [-Wshadow]
  852 |  void *__mptr = (void *)(ptr);     \
      |        ^~~~~~

Rewrite the macro to use an inline function internally, which makes it
more readable and reduces the amount of useless output from make W=2.

Fixes: 730623f4a56f ("IB/verbs: Add helper function rdma_udata_to_drv_context")
Link: https://lore.kernel.org/r/20201026161549.3709175-1-arnd@kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/bnxt_re: Fix entry size during SRQ create
Selvin Xavier [Tue, 13 Oct 2020 06:15:52 +0000 (23:15 -0700)]
RDMA/bnxt_re: Fix entry size during SRQ create

Only static WQE is supported for SRQ. So always use the max supported SGEs
while calculating SRQ entry size.

Fixes: 2bb3c32c5c5f ("RDMA/bnxt_re: Change wr posting logic to accommodate variable wqes")
Link: https://lore.kernel.org/r/1602569752-12745-1-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/rxe: Remove unused RXE_MR_TYPE_FMR
Bob Pearson [Fri, 9 Oct 2020 16:51:13 +0000 (11:51 -0500)]
RDMA/rxe: Remove unused RXE_MR_TYPE_FMR

This is a left over from the past. It is no longer used.

Link: https://lore.kernel.org/r/20201009165112.271143-1-rpearson@hpe.com
Signed-off-by: Bob Pearson <rpearson@hpe.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Convert sysfs kobject * show functions to use sysfs_emit()
Joe Perches [Thu, 8 Oct 2020 02:36:25 +0000 (19:36 -0700)]
RDMA: Convert sysfs kobject * show functions to use sysfs_emit()

Done with cocci script:

@@
identifier k_show;
identifier arg1, arg2, arg3;
@@
ssize_t k_show(struct kobject *
- arg1
+ kobj
, struct kobj_attribute *
- arg2
+ attr
, char *
- arg3
+ buf
)
{
...
(
- arg1
+ kobj
|
- arg2
+ attr
|
- arg3
+ buf
)
...
}

@@
identifier k_show;
identifier kobj, attr, buf;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
return
- sprintf(buf,
+ sysfs_emit(buf,
...);
...>
}

@@
identifier k_show;
identifier kobj, attr, buf;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
return
- snprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
}

@@
identifier k_show;
identifier kobj, attr, buf;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
return
- scnprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
}

@@
identifier k_show;
identifier kobj, attr, buf;
expression chr;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
return
- strcpy(buf, chr);
+ sysfs_emit(buf, chr);
...>
}

@@
identifier k_show;
identifier kobj, attr, buf;
identifier len;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
len =
- sprintf(buf,
+ sysfs_emit(buf,
...);
...>
return len;
}

@@
identifier k_show;
identifier kobj, attr, buf;
identifier len;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
len =
- snprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
return len;
}

@@
identifier k_show;
identifier kobj, attr, buf;
identifier len;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
len =
- scnprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
return len;
}

@@
identifier k_show;
identifier kobj, attr, buf;
identifier len;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
<...
- len += scnprintf(buf + len, PAGE_SIZE - len,
+ len += sysfs_emit_at(buf, len,
...);
...>
return len;
}

@@
identifier k_show;
identifier kobj, attr, buf;
expression chr;
@@

ssize_t k_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
{
...
- strcpy(buf, chr);
- return strlen(buf);
+ return sysfs_emit(buf, chr);
}

Link: https://lore.kernel.org/r/7761c1efaebb96c432c85171d58405c25a824ccd.1602122880.git.joe@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Convert sysfs device * show functions to use sysfs_emit()
Joe Perches [Thu, 8 Oct 2020 02:36:24 +0000 (19:36 -0700)]
RDMA: Convert sysfs device * show functions to use sysfs_emit()

Done with cocci script:

@@
identifier d_show;
identifier dev, attr, buf;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
return
- sprintf(buf,
+ sysfs_emit(buf,
...);
...>
}

@@
identifier d_show;
identifier dev, attr, buf;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
return
- snprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
}

@@
identifier d_show;
identifier dev, attr, buf;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
return
- scnprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
}

@@
identifier d_show;
identifier dev, attr, buf;
expression chr;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
return
- strcpy(buf, chr);
+ sysfs_emit(buf, chr);
...>
}

@@
identifier d_show;
identifier dev, attr, buf;
identifier len;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
len =
- sprintf(buf,
+ sysfs_emit(buf,
...);
...>
return len;
}

@@
identifier d_show;
identifier dev, attr, buf;
identifier len;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
len =
- snprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
return len;
}

@@
identifier d_show;
identifier dev, attr, buf;
identifier len;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
len =
- scnprintf(buf, PAGE_SIZE,
+ sysfs_emit(buf,
...);
...>
return len;
}

@@
identifier d_show;
identifier dev, attr, buf;
identifier len;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
<...
- len += scnprintf(buf + len, PAGE_SIZE - len,
+ len += sysfs_emit_at(buf, len,
...);
...>
return len;
}

@@
identifier d_show;
identifier dev, attr, buf;
expression chr;
@@

ssize_t d_show(struct device *dev, struct device_attribute *attr, char *buf)
{
...
- strcpy(buf, chr);
- return strlen(buf);
+ return sysfs_emit(buf, chr);
}

Link: https://lore.kernel.org/r/7f406fa8e3aa2552c022bec680f621e38d1fe414.1602122879.git.joe@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Remove AH from uverbs_cmd_mask
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:11 +0000 (20:20 -0300)]
RDMA: Remove AH from uverbs_cmd_mask

Drivers that need a uverbs AH should instead set the create_user_ah() op
similar to reg_user_mr(). MODIFY_AH and QUERY_AH cmds were never
implemented so are just deleted.

Link: https://lore.kernel.org/r/11-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Remove uverbs cmds from drivers that don't use them
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:10 +0000 (20:20 -0300)]
RDMA: Remove uverbs cmds from drivers that don't use them

Allowing userspace to invoke these commands is probably going to crash
these drivers as they are not tested and not expecting to use them on a
user object.

For example pvrdma touches cq->ring_state which is not initialized for
user QPs.

These commands are effected:

- IB_USER_VERBS_CMD_REQ_NOTIFY_CQ is ibv_cmd_req_notify_cq() in
  rdma-core, only hfi1, ipath and rxe calls it.

- IB_USER_VERBS_CMD_POLL_CQ is ibv_cmd_poll_cq() in rdma-core, only
  ipath and hfi1 calls it.

- IB_USER_VERBS_CMD_POST_SEND/RECV is ibv_cmd_post_send/recv() in
  rdma-core, only ipath and hfi1 call them.

- IB_USER_VERBS_CMD_POST_SRQ_RECV is ibv_cmd_post_srq_recv() in
  rdma-core, only ipath and hfi1 calls it.

- IB_USER_VERBS_CMD_PEEK_CQ isn't even implemented anywhere

- IB_USER_VERBS_CMD_CREATE/DESTROY_AH is ibv_cmd_create/destroy_ah() in
  rdma-core, only bnxt_re, efa, hfi1, ipath, mlx5, orcrdma, and rxe call
  it.

Link: https://lore.kernel.org/r/10-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/core Remove uverbs_ex_cmd_mask
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:09 +0000 (20:20 -0300)]
RDMA/core Remove uverbs_ex_cmd_mask

No driver sets it, and the core code sets a maximum mask, simply remove
it.

Disabled operations are now handled either by having a NULL ops pointer,
or by having the common driver callbacks check for unsupported extended
attributes.

Link: https://lore.kernel.org/r/9-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Check create_flags during create_qp
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:08 +0000 (20:20 -0300)]
RDMA: Check create_flags during create_qp

Each driver should check that the QP attrs create_flags is supported.
Unfortuantely when create_flags was added to the QP attrs the drivers were
not updated. uverbs_ex_cmd_mask was used to block it - even though kernel
drivers use these flags too.

Check that flags is zero in all drivers that don't use it, remove
IB_USER_VERBS_EX_CMD_CREATE_QP from uverbs_ex_cmd_mask. Fix the error code
to be EOPNOTSUPP.

Link: https://lore.kernel.org/r/8-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Check flags during create_cq
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:07 +0000 (20:20 -0300)]
RDMA: Check flags during create_cq

Each driver should check that the CQ attrs is supported. Unfortuantely
when flags was added to the CQ attrs the drivers were not updated,
uverbs_ex_cmd_mask was used to block it. This was missed when create CQ
was converted to ioctl, so non-zero flags could have been passed into
drivers.

Check that flags is zero in all drivers that don't use it, remove
IB_USER_VERBS_EX_CMD_CREATE_CQ from uverbs_ex_cmd_mask.

Fixes: 41b2a71fc848 ("IB/uverbs: Move ioctl path of create_cq and destroy_cq to a new file")
Link: https://lore.kernel.org/r/7-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Check attr_mask during modify_qp
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:06 +0000 (20:20 -0300)]
RDMA: Check attr_mask during modify_qp

Each driver should check that it can support the provided attr_mask during
modify_qp. IB_USER_VERBS_EX_CMD_MODIFY_QP was being used to block
modify_qp_ex because the driver didn't check RATE_LIMIT.

Link: https://lore.kernel.org/r/6-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Check srq_type during create_srq
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:05 +0000 (20:20 -0300)]
RDMA: Check srq_type during create_srq

uverbs was blocking srq_types the driver doesn't support based on the
CREATE_XSRQ cmd_mask. Fix all drivers to check for supported srq_types
during create_srq and move CREATE_XSRQ to the core code.

Link: https://lore.kernel.org/r/5-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Move more uverbs_cmd_mask settings to the core
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:04 +0000 (20:20 -0300)]
RDMA: Move more uverbs_cmd_mask settings to the core

These functions all depend on the driver providing a specific op:

- REREG_MR is rereg_user_mr(). bnxt_re set this without providing the op
- ATTACH/DEATCH_MCAST is attach_mcast()/detach_mcast(). usnic set this
  without providing the op
- OPEN_QP doesn't involve the driver but requires a XRCD. qedr provides
  xrcd but forgot to set it, usnic doesn't provide XRCD but set it anyhow.
- OPEN/CLOSE_XRCD are the ops alloc_xrcd()/dealloc_xrcd()
- CREATE_SRQ/DESTROY_SRQ are the ops create_srq()/destroy_srq()
- QUERY/MODIFY_SRQ is op query_srq()/modify_srq(). hns sets this but
  sometimes supplies a NULL op.
- RESIZE_CQ is op resize_cq(). bnxt_re sets this boes doesn't supply an op
- ALLOC/DEALLOC_MW is alloc_mw()/dealloc_mw(). cxgb4 provided an
  (now deleted) implementation but no userspace

All drivers were checked that no drivers provide the op without also
setting uverbs_cmd_mask so this should have no functional change.

Link: https://lore.kernel.org/r/4-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Remove elements in uverbs_cmd_mask that all drivers set
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:03 +0000 (20:20 -0300)]
RDMA: Remove elements in uverbs_cmd_mask that all drivers set

This is a step toward eliminating uverbs_cmd_mask. Preset this list in the
core code. Only the op reg_user_mr wasn't already being required from the
drivers.

Link: https://lore.kernel.org/r/3-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA: Remove uverbs_ex_cmd_mask values that are linked to functions
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:02 +0000 (20:20 -0300)]
RDMA: Remove uverbs_ex_cmd_mask values that are linked to functions

Since a while now the uverbs layer checks if the driver implements a
function before allowing the ucmd to proceed. This largely obsoletes the
cmd_mask stuff, but there is some tricky bits in drivers preventing it
from being removed.

Remove the easy elements of uverbs_ex_cmd_mask by pre-setting them in the
core code. These are triggered soley based on the related ops function
pointer.

query_device_ex is not triggered based on an op, but all drivers already
implement something compatible with the extension, so enable it globally
too.

Link: https://lore.kernel.org/r/2-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/cxgb4: Remove MW support
Jason Gunthorpe [Sat, 3 Oct 2020 23:20:01 +0000 (20:20 -0300)]
RDMA/cxgb4: Remove MW support

This driver never enabled IB_USER_VERBS_CMD_ALLOC_MW so memory windows
were not usable from userspace. The kernel side was removed long ago. Drop
this dead code.

Fixes: feb7c1e38bcc ("IB: remove in-kernel support for memory windows")
Link: https://lore.kernel.org/r/1-v1-caa70ba3d1ab+1436e-ucmd_mask_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoRDMA/bnxt_re: Set queue pair state when being queried
Kamal Heib [Wed, 21 Oct 2020 11:49:52 +0000 (14:49 +0300)]
RDMA/bnxt_re: Set queue pair state when being queried

The API for ib_query_qp requires the driver to set cur_qp_state on return,
add the missing set.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/20201021114952.38876-1-kamalheib1@gmail.com
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
4 years agoLinux 5.10-rc1
Linus Torvalds [Sun, 25 Oct 2020 22:14:11 +0000 (15:14 -0700)]
Linux 5.10-rc1

4 years agotreewide: Convert macro and uses of __section(foo) to __section("foo")
Joe Perches [Thu, 22 Oct 2020 02:36:07 +0000 (19:36 -0700)]
treewide: Convert macro and uses of __section(foo) to __section("foo")

Use a more generic form for __section that requires quotes to avoid
complications with clang and gcc differences.

Remove the quote operator # from compiler_attributes.h __section macro.

Convert all unquoted __section(foo) uses to quoted __section("foo").
Also convert __attribute__((section("foo"))) uses to __section("foo")
even if the __attribute__ has multiple list entry forms.

Conversion done using the script at:

    https://lore.kernel.org/lkml/75393e5ddc272dc7403de74d645e6c6e0f4e70eb.camel@perches.com/2-convert_section.pl

Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@gooogle.com>
Reviewed-by: Miguel Ojeda <ojeda@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agokernel/sys.c: fix prototype of prctl_get_tid_address()
Rasmus Villemoes [Sat, 24 Oct 2020 01:04:26 +0000 (03:04 +0200)]
kernel/sys.c: fix prototype of prctl_get_tid_address()

tid_addr is not a "pointer to (pointer to int in userspace)"; it is in
fact a "pointer to (pointer to int in userspace) in userspace".  So
sparse rightfully complains about passing a kernel pointer to
put_user().

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agomm: remove kzfree() compatibility definition
Eric Biggers [Fri, 23 Oct 2020 23:27:16 +0000 (16:27 -0700)]
mm: remove kzfree() compatibility definition

Commit 453431a54934 ("mm, treewide: rename kzfree() to
kfree_sensitive()") renamed kzfree() to kfree_sensitive(),
but it left a compatibility definition of kzfree() to avoid
being too disruptive.

Since then a few more instances of kzfree() have slipped in.

Just get rid of them and remove the compatibility definition
once and for all.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agocheckpatch: enable GIT_DIR environment use to set git repository location
Joe Perches [Sat, 24 Oct 2020 23:59:04 +0000 (16:59 -0700)]
checkpatch: enable GIT_DIR environment use to set git repository location

If set, use the environment variable GIT_DIR to change the default .git
location of the kernel git tree.

If GIT_DIR is unset, keep using the current ".git" default.

Link: https://lkml.kernel.org/r/c5e23b45562373d632fccb8bc04e563abba4dd1d.camel@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
4 years agoMerge tag 'timers-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 25 Oct 2020 18:28:49 +0000 (11:28 -0700)]
Merge tag 'timers-urgent-2020-10-25' of git://git./linux/kernel/git/tip/tip

Pull timer fixes from Thomas Gleixner:
 "A time namespace fix and a matching selftest. The futex absolute
  timeouts which are based on CLOCK_MONOTONIC require time namespace
  corrected. This was missed in the original time namesapce support"

* tag 'timers-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  selftests/timens: Add a test for futex()
  futex: Adjust absolute futex timeouts with per time namespace offset

4 years agoMerge tag 'sched-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 25 Oct 2020 18:25:16 +0000 (11:25 -0700)]
Merge tag 'sched-urgent-2020-10-25' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Thomas Gleixner:
 "Two scheduler fixes:

   - A trivial build fix for sched_feat() to compile correctly with
     CONFIG_JUMP_LABEL=n

   - Replace a zero lenght array with a flexible array"

* tag 'sched-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/features: Fix !CONFIG_JUMP_LABEL case
  sched: Replace zero-length array with flexible-array

4 years agoMerge tag 'perf-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 25 Oct 2020 18:22:59 +0000 (11:22 -0700)]
Merge tag 'perf-urgent-2020-10-25' of git://git./linux/kernel/git/tip/tip

Pull perf fix from Thomas Gleixner:
 "A single fix to compute the field offset of the SNOOPX bit in the data
  source bitmask of perf events correctly"

* tag 'perf-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf: correct SNOOPX field offset

4 years agoMerge tag 'locking-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 25 Oct 2020 18:14:54 +0000 (11:14 -0700)]
Merge tag 'locking-urgent-2020-10-25' of git://git./linux/kernel/git/tip/tip

Pull locking fix from Thomas Gleixner:
 "Just a trivial fix for kernel-doc warnings"

* tag 'locking-urgent-2020-10-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/seqlocks: Fix kernel-doc warnings

4 years agoMerge tag 'ntb-5.10' of git://github.com/jonmason/ntb
Linus Torvalds [Sun, 25 Oct 2020 18:12:31 +0000 (11:12 -0700)]
Merge tag 'ntb-5.10' of git://github.com/jonmason/ntb

Pull NTB fixes from Jon Mason.

* tag 'ntb-5.10' of git://github.com/jonmason/ntb:
  NTB: Use struct_size() helper in devm_kzalloc()
  ntb: intel: Fix memleak in intel_ntb_pci_probe
  NTB: hw: amd: fix an issue about leak system resources

4 years agoMerge branch 'i2c/for-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sun, 25 Oct 2020 18:10:23 +0000 (11:10 -0700)]
Merge branch 'i2c/for-5.10' of git://git./linux/kernel/git/wsa/linux

Pull i2c fix from Wolfram Sang:
 "Regression fix for rc1 and stable kernels as well"

* 'i2c/for-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: core: Restore acpi_walk_dep_device_list() getting called after registering the ACPI i2c devs

4 years agoMerge tag '5.10-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sun, 25 Oct 2020 18:05:04 +0000 (11:05 -0700)]
Merge tag '5.10-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6

Pull more cifs updates from Steve French:
 "Add support for stat of various special file types (WSL reparse points
  for char, block, fifo)"

* tag '5.10-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: update internal module version number
  smb3: add some missing definitions from MS-FSCC
  smb3: remove two unused variables
  smb3: add support for stat of WSL reparse points for special file types

4 years agoMerge branch 'parisc-5.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sun, 25 Oct 2020 17:59:34 +0000 (10:59 -0700)]
Merge branch 'parisc-5.10-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull more parisc updates from Helge Deller:

 - During this merge window O_NONBLOCK was changed to become 000200000,
   but we missed that the syscalls timerfd_create(), signalfd4(),
   eventfd2(), pipe2(), inotify_init1() and userfaultfd() do a strict
   bit-wise check of the flags parameter.

   To provide backward compatibility with existing userspace we
   introduce parisc specific wrappers for those syscalls which filter
   out the old O_NONBLOCK value and replaces it with the new one.

 - Prevent HIL bus driver to get stuck when keyboard or mouse isn't
   attached

 - Improve error return codes when setting rtc time

 - Minor documentation fix in pata_ns87415.c

* 'parisc-5.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  ata: pata_ns87415.c: Document support on parisc with superio chip
  parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
  hil/parisc: Disable HIL driver when it gets stuck
  parisc: Improve error return codes when setting rtc time

4 years agoMerge tag 'for-linus-5.10b-rc1c-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 25 Oct 2020 17:55:35 +0000 (10:55 -0700)]
Merge tag 'for-linus-5.10b-rc1c-tag' of git://git./linux/kernel/git/xen/tip

Pull more xen updates from Juergen Gross:

 - a series for the Xen pv block drivers adding module parameters for
   better control of resource usge

 - a cleanup series for the Xen event driver

* tag 'for-linus-5.10b-rc1c-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  Documentation: add xen.fifo_events kernel parameter description
  xen/events: unmask a fifo event channel only if it was masked
  xen/events: only register debug interrupt for 2-level events
  xen/events: make struct irq_info private to events_base.c
  xen: remove no longer used functions
  xen-blkfront: Apply changed parameter name to the document
  xen-blkfront: add a parameter for disabling of persistent grants
  xen-blkback: add a parameter for disabling of persistent grants

4 years agoMerge tag 'safesetid-5.10' of git://github.com/micah-morton/linux
Linus Torvalds [Sun, 25 Oct 2020 17:45:26 +0000 (10:45 -0700)]
Merge tag 'safesetid-5.10' of git://github.com/micah-morton/linux

Pull SafeSetID updates from Micah Morton:
 "The changes are mostly contained to within the SafeSetID LSM, with the
  exception of a few 1-line changes to change some ns_capable() calls to
  ns_capable_setid() -- causing a flag (CAP_OPT_INSETID) to be set that
  is examined by SafeSetID code and nothing else in the kernel.

  The changes to SafeSetID internally allow for setting up GID
  transition security policies, as already existed for UIDs"

* tag 'safesetid-5.10' of git://github.com/micah-morton/linux:
  LSM: SafeSetID: Fix warnings reported by test bot
  LSM: SafeSetID: Add GID security policy handling
  LSM: Signal to SafeSetID when setting group IDs

4 years agoMerge tag '20201024-v4-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/wtarrea...
Linus Torvalds [Sun, 25 Oct 2020 17:40:08 +0000 (10:40 -0700)]
Merge tag '20201024-v4-5.10' of git://git./linux/kernel/git/wtarreau/prandom

Pull random32 updates from Willy Tarreau:
 "Make prandom_u32() less predictable.

  This is the cleanup of the latest series of prandom_u32
  experimentations consisting in using SipHash instead of Tausworthe to
  produce the randoms used by the network stack.

  The changes to the files were kept minimal, and the controversial
  commit that used to take noise from the fast_pool (f227e3ec3b5c) was
  reverted. Instead, a dedicated "net_rand_noise" per_cpu variable is
  fed from various sources of activities (networking, scheduling) to
  perturb the SipHash state using fast, non-trivially predictable data,
  instead of keeping it fully deterministic. The goal is essentially to
  make any occasional memory leakage or brute-force attempt useless.

  The resulting code was verified to be very slightly faster on x86_64
  than what is was with the controversial commit above, though this
  remains barely above measurement noise. It was also tested on i386 and
  arm, and build- tested only on arm64"

Link: https://lore.kernel.org/netdev/20200808152628.GA27941@SDF.ORG/
* tag '20201024-v4-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/wtarreau/prandom:
  random32: add a selftest for the prandom32 code
  random32: add noise from network and scheduling activity
  random32: make prandom_u32() output unpredictable

4 years agoi2c: core: Restore acpi_walk_dep_device_list() getting called after registering the...
Hans de Goede [Wed, 14 Oct 2020 14:41:58 +0000 (16:41 +0200)]
i2c: core: Restore acpi_walk_dep_device_list() getting called after registering the ACPI i2c devs

Commit 21653a4181ff ("i2c: core: Call i2c_acpi_install_space_handler()
before i2c_acpi_register_devices()")'s intention was to only move the
acpi_install_address_space_handler() call to the point before where
the ACPI declared i2c-children of the adapter where instantiated by
i2c_acpi_register_devices().

But i2c_acpi_install_space_handler() had a call to
acpi_walk_dep_device_list() hidden (that is I missed it) at the end
of it, so as an unwanted side-effect now acpi_walk_dep_device_list()
was also being called before i2c_acpi_register_devices().

Move the acpi_walk_dep_device_list() call to the end of
i2c_acpi_register_devices(), so that it is once again called *after*
the i2c_client-s hanging of the adapter have been created.

This fixes the Microsoft Surface Go 2 hanging at boot.

Fixes: 21653a4181ff ("i2c: core: Call i2c_acpi_install_space_handler() before i2c_acpi_register_devices()")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=209627
Reported-by: Rainer Finke <rainer@finke.cc>
Reported-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Suggested-by: Maximilian Luz <luzmaximilian@gmail.com>
Tested-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Wolfram Sang <wsa@kernel.org>
4 years agoMerge tag 'block-5.10-2020-10-24' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 24 Oct 2020 19:46:42 +0000 (12:46 -0700)]
Merge tag 'block-5.10-2020-10-24' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - NVMe pull request from Christoph
     - rdma error handling fixes (Chao Leng)
     - fc error handling and reconnect fixes (James Smart)
     - fix the qid displace when tracing ioctl command (Keith Busch)
     - don't use BLK_MQ_REQ_NOWAIT for passthru (Chaitanya Kulkarni)
     - fix MTDT for passthru (Logan Gunthorpe)
     - blacklist Write Same on more devices (Kai-Heng Feng)
     - fix an uninitialized work struct (zhenwei pi)"

 - lightnvm out-of-bounds fix (Colin)

 - SG allocation leak fix (Doug)

 - rnbd fixes (Gioh, Guoqing, Jack)

 - zone error translation fixes (Keith)

 - kerneldoc markup fix (Mauro)

 - zram lockdep fix (Peter)

 - Kill unused io_context members (Yufen)

 - NUMA memory allocation cleanup (Xianting)

 - NBD config wakeup fix (Xiubo)

* tag 'block-5.10-2020-10-24' of git://git.kernel.dk/linux-block: (27 commits)
  block: blk-mq: fix a kernel-doc markup
  nvme-fc: shorten reconnect delay if possible for FC
  nvme-fc: wait for queues to freeze before calling update_hr_hw_queues
  nvme-fc: fix error loop in create_hw_io_queues
  nvme-fc: fix io timeout to abort I/O
  null_blk: use zone status for max active/open
  nvmet: don't use BLK_MQ_REQ_NOWAIT for passthru
  nvmet: cleanup nvmet_passthru_map_sg()
  nvmet: limit passthru MTDS by BIO_MAX_PAGES
  nvmet: fix uninitialized work for zero kato
  nvme-pci: disable Write Zeroes on Sandisk Skyhawk
  nvme: use queuedata for nvme_req_qid
  nvme-rdma: fix crash due to incorrect cqe
  nvme-rdma: fix crash when connect rejected
  block: remove unused members for io_context
  blk-mq: remove the calling of local_memory_node()
  zram: Fix __zram_bvec_{read,write}() locking order
  skd_main: remove unused including <linux/version.h>
  sgl_alloc_order: fix memory leak
  lightnvm: fix out-of-bounds write to array devices->info[]
  ...

4 years agoMerge tag 'io_uring-5.10-2020-10-24' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 24 Oct 2020 19:40:18 +0000 (12:40 -0700)]
Merge tag 'io_uring-5.10-2020-10-24' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:

 - fsize was missed in previous unification of work flags

 - Few fixes cleaning up the flags unification creds cases (Pavel)

 - Fix NUMA affinities for completely unplugged/replugged node for io-wq

 - Two fallout fixes from the set_fs changes. One local to io_uring, one
   for the splice entry point that io_uring uses.

 - Linked timeout fixes (Pavel)

 - Removal of ->flush() ->files work-around that we don't need anymore
   with referenced files (Pavel)

 - Various cleanups (Pavel)

* tag 'io_uring-5.10-2020-10-24' of git://git.kernel.dk/linux-block:
  splice: change exported internal do_splice() helper to take kernel offset
  io_uring: make loop_rw_iter() use original user supplied pointers
  io_uring: remove req cancel in ->flush()
  io-wq: re-set NUMA node affinities if CPUs come online
  io_uring: don't reuse linked_timeout
  io_uring: unify fsize with def->work_flags
  io_uring: fix racy REQ_F_LINK_TIMEOUT clearing
  io_uring: do poll's hash_node init in common code
  io_uring: inline io_poll_task_handler()
  io_uring: remove extra ->file check in poll prep
  io_uring: make cached_cq_overflow non atomic_t
  io_uring: inline io_fail_links()
  io_uring: kill ref get/drop in personality init
  io_uring: flags-based creds init in queue

4 years agoMerge tag 'libata-5.10-2020-10-24' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 24 Oct 2020 19:36:24 +0000 (12:36 -0700)]
Merge tag 'libata-5.10-2020-10-24' of git://git.kernel.dk/linux-block

Pull libata fixes from Jens Axboe:
 "Two minor libata fixes:

   - Fix a DMA boundary mask regression for sata_rcar (Geert)

   - kerneldoc markup fix (Mauro)"

* tag 'libata-5.10-2020-10-24' of git://git.kernel.dk/linux-block:
  ata: fix some kernel-doc markups
  ata: sata_rcar: Fix DMA boundary mask

4 years agoMerge branch 'work.misc' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sat, 24 Oct 2020 19:26:05 +0000 (12:26 -0700)]
Merge branch 'work.misc' of git://git./linux/kernel/git/viro/vfs

Pull misc vfs updates from Al Viro:
 "Assorted stuff all over the place (the largest group here is
  Christoph's stat cleanups)"

* 'work.misc' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  fs: remove KSTAT_QUERY_FLAGS
  fs: remove vfs_stat_set_lookup_flags
  fs: move vfs_fstatat out of line
  fs: implement vfs_stat and vfs_lstat in terms of vfs_fstatat
  fs: remove vfs_statx_fd
  fs: omfs: use kmemdup() rather than kmalloc+memcpy
  [PATCH] reduce boilerplate in fsid handling
  fs: Remove duplicated flag O_NDELAY occurring twice in VALID_OPEN_FLAGS
  selftests: mount: add nosymfollow tests
  Add a "nosymfollow" mount option.

4 years agoMerge tag 'dma-mapping-5.10-1' of git://git.infradead.org/users/hch/dma-mapping
Linus Torvalds [Sat, 24 Oct 2020 19:17:05 +0000 (12:17 -0700)]
Merge tag 'dma-mapping-5.10-1' of git://git.infradead.org/users/hch/dma-mapping

Pull dma-mapping fixes from Christoph Hellwig:

 - document the new dma_{alloc,free}_pages() API

 - two fixups for the dma-mapping.h split

* tag 'dma-mapping-5.10-1' of git://git.infradead.org/users/hch/dma-mapping:
  dma-mapping: document dma_{alloc,free}_pages
  dma-mapping: move more functions to dma-map-ops.h
  ARM/sa1111: add a missing include of dma-map-ops.h

4 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sat, 24 Oct 2020 19:09:22 +0000 (12:09 -0700)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "Two fixes for this merge window, and an unrelated bugfix for a host
  hang"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: ioapic: break infinite recursion on lazy EOI
  KVM: vmx: rename pi_init to avoid conflict with paride
  KVM: x86/mmu: Avoid modulo operator on 64-bit value to fix i386 build

4 years agoMerge tag 'x86_seves_fixes_for_v5.10_rc1' of git://git.kernel.org/pub/scm/linux/kerne...
Linus Torvalds [Sat, 24 Oct 2020 18:49:32 +0000 (11:49 -0700)]
Merge tag 'x86_seves_fixes_for_v5.10_rc1' of git://git./linux/kernel/git/tip/tip

Pull x86 SEV-ES fixes from Borislav Petkov:
 "Three fixes to SEV-ES to correct setting up the new early pagetable on
  5-level paging machines, to always map boot_params and the kernel
  cmdline, and disable stack protector for ../compressed/head{32,64}.c.
  (Arvind Sankar)"

* tag 'x86_seves_fixes_for_v5.10_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/boot/64: Explicitly map boot_params and command line
  x86/head/64: Disable stack protection for head$(BITS).o
  x86/boot/64: Initialize 5-level paging variables earlier

4 years agorandom32: add a selftest for the prandom32 code
Willy Tarreau [Sat, 24 Oct 2020 16:36:27 +0000 (18:36 +0200)]
random32: add a selftest for the prandom32 code

Given that this code is new, let's add a selftest for it as well.
It doesn't rely on fixed sets, instead it picks 1024 numbers and
verifies that they're not more correlated than desired.

Link: https://lore.kernel.org/netdev/20200808152628.GA27941@SDF.ORG/
Cc: George Spelvin <lkml@sdf.org>
Cc: Amit Klein <aksecurity@gmail.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: tytso@mit.edu
Cc: Florian Westphal <fw@strlen.de>
Cc: Marc Plumb <lkml.mplumb@gmail.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
4 years agorandom32: add noise from network and scheduling activity
Willy Tarreau [Mon, 10 Aug 2020 08:27:42 +0000 (10:27 +0200)]
random32: add noise from network and scheduling activity

With the removal of the interrupt perturbations in previous random32
change (random32: make prandom_u32() output unpredictable), the PRNG
has become 100% deterministic again. While SipHash is expected to be
way more robust against brute force than the previous Tausworthe LFSR,
there's still the risk that whoever has even one temporary access to
the PRNG's internal state is able to predict all subsequent draws till
the next reseed (roughly every minute). This may happen through a side
channel attack or any data leak.

This patch restores the spirit of commit f227e3ec3b5c ("random32: update
the net random state on interrupt and activity") in that it will perturb
the internal PRNG's statee using externally collected noise, except that
it will not pick that noise from the random pool's bits nor upon
interrupt, but will rather combine a few elements along the Tx path
that are collectively hard to predict, such as dev, skb and txq
pointers, packet length and jiffies values. These ones are combined
using a single round of SipHash into a single long variable that is
mixed with the net_rand_state upon each invocation.

The operation was inlined because it produces very small and efficient
code, typically 3 xor, 2 add and 2 rol. The performance was measured
to be the same (even very slightly better) than before the switch to
SipHash; on a 6-core 12-thread Core i7-8700k equipped with a 40G NIC
(i40e), the connection rate dropped from 556k/s to 555k/s while the
SYN cookie rate grew from 5.38 Mpps to 5.45 Mpps.

Link: https://lore.kernel.org/netdev/20200808152628.GA27941@SDF.ORG/
Cc: George Spelvin <lkml@sdf.org>
Cc: Amit Klein <aksecurity@gmail.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: tytso@mit.edu
Cc: Florian Westphal <fw@strlen.de>
Cc: Marc Plumb <lkml.mplumb@gmail.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
4 years agorandom32: make prandom_u32() output unpredictable
George Spelvin [Sun, 9 Aug 2020 06:57:44 +0000 (06:57 +0000)]
random32: make prandom_u32() output unpredictable

Non-cryptographic PRNGs may have great statistical properties, but
are usually trivially predictable to someone who knows the algorithm,
given a small sample of their output.  An LFSR like prandom_u32() is
particularly simple, even if the sample is widely scattered bits.

It turns out the network stack uses prandom_u32() for some things like
random port numbers which it would prefer are *not* trivially predictable.
Predictability led to a practical DNS spoofing attack.  Oops.

This patch replaces the LFSR with a homebrew cryptographic PRNG based
on the SipHash round function, which is in turn seeded with 128 bits
of strong random key.  (The authors of SipHash have *not* been consulted
about this abuse of their algorithm.)  Speed is prioritized over security;
attacks are rare, while performance is always wanted.

Replacing all callers of prandom_u32() is the quick fix.
Whether to reinstate a weaker PRNG for uses which can tolerate it
is an open question.

Commit f227e3ec3b5c ("random32: update the net random state on interrupt
and activity") was an earlier attempt at a solution.  This patch replaces
it.

Reported-by: Amit Klein <aksecurity@gmail.com>
Cc: Willy Tarreau <w@1wt.eu>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: tytso@mit.edu
Cc: Florian Westphal <fw@strlen.de>
Cc: Marc Plumb <lkml.mplumb@gmail.com>
Fixes: f227e3ec3b5c ("random32: update the net random state on interrupt and activity")
Signed-off-by: George Spelvin <lkml@sdf.org>
Link: https://lore.kernel.org/netdev/20200808152628.GA27941@SDF.ORG/
[ willy: partial reversal of f227e3ec3b5c; moved SIPROUND definitions
  to prandom.h for later use; merged George's prandom_seed() proposal;
  inlined siprand_u32(); replaced the net_rand_state[] array with 4
  members to fix a build issue; cosmetic cleanups to make checkpatch
  happy; fixed RANDOM32_SELFTEST build ]
Signed-off-by: Willy Tarreau <w@1wt.eu>