linux.git/drivers/infiniband, branch v5.5-rc2

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

2019-12-15T22:58:13+00:00

Pull rdma fixes from Doug Ledford:
 "A small collection of -rc fixes. Mostly. One API addition, but that's
  because we wanted to use it in a fix. There's also a bug fix that is
  going to render the 5.5 kernel's soft-RoCE driver incompatible with
  all soft-RoCE versions prior, but it's required to actually implement
  the protocol according to the RoCE spec and required in order for the
  soft-RoCE driver to be able to successfully work with actual RoCE
  hardware.

  Summary:

   - Update Steve Wise info

   - Fix for soft-RoCE crc calculations (will break back compatibility,
     but only with the soft-RoCE driver, which has had this bug since it
     was introduced and it is an on-the-wire bug, but will make
     soft-RoCE fully compatible with real RoCE hardware)

   - cma init fixup

   - counters oops fix

   - fix for mlx4 init/teardown sequence

   - fix for mkx5 steering rules

   - introduce a cleanup API, which isn't a fix, but we want to use it
     in the next fix

   - fix for mlx5 memory management that uses API in previous patch"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  IB/mlx5: Fix device memory flows
  IB/core: Introduce rdma_user_mmap_entry_insert_range() API
  IB/mlx5: Fix steering rule of drop and count
  IB/mlx4: Follow mirror sequence of device add during device removal
  RDMA/counter: Prevent auto-binding a QP which are not tracked with res
  rxe: correctly calculate iCRC for unaligned payloads
  Update mailmap info for Steve Wise
  RDMA/cma: add missed unregister_pernet_subsys in init failure

IB/mlx5: Fix device memory flows

2019-12-12T21:55:36+00:00

Fix device memory flows so that only once there will be no live mmaped
VA to a given allocation the matching object will be destroyed.

This prevents a potential scenario that existing VA that was mmaped by
one process might still be used post its deallocation despite that it's
owned now by other process.

The above is achieved by integrating with IB core APIs to manage
mmap/munmap. Only once the refcount will become 0 the DM object and its
underlay area will be freed.

Fixes: 3b113a1ec3d4 ("IB/mlx5: Support device memory type attribute")
Signed-off-by: Yishai Hadas 
Signed-off-by: Leon Romanovsky 
Link: https://lore.kernel.org/r/20191212100237.330654-3-leon@kernel.org
Signed-off-by: Doug Ledford

IB/core: Introduce rdma_user_mmap_entry_insert_range() API

2019-12-12T21:55:36+00:00

Introduce rdma_user_mmap_entry_insert_range() API to be used once the
required key for the given entry should be in a given range.

Signed-off-by: Yishai Hadas 
Signed-off-by: Leon Romanovsky 
Link: https://lore.kernel.org/r/20191212100237.330654-2-leon@kernel.org
Signed-off-by: Doug Ledford

IB/mlx5: Fix steering rule of drop and count

2019-12-12T20:38:16+00:00

There are two flow rule destinations: QP and packet. While users are
setting DROP packet rule, the QP should not be set as a destination.

Fixes: 3b3233fbf02e ("IB/mlx5: Add flow counters binding support")
Signed-off-by: Maor Gottlieb 
Reviewed-by: Raed Salem 
Signed-off-by: Leon Romanovsky 
Link: https://lore.kernel.org/r/20191212091214.315005-4-leon@kernel.org
Signed-off-by: Doug Ledford

IB/mlx4: Follow mirror sequence of device add during device removal

2019-12-12T20:38:15+00:00

Current code device add sequence is:

ib_register_device()
ib_mad_init()
init_sriov_init()
register_netdev_notifier()

Therefore, the remove sequence should be,

unregister_netdev_notifier()
close_sriov()
mad_cleanup()
ib_unregister_device()

However it is not above.
Hence, make do above remove sequence.

Fixes: fa417f7b520ee ("IB/mlx4: Add support for IBoE")
Signed-off-by: Parav Pandit 
Reviewed-by: Maor Gottlieb 
Signed-off-by: Leon Romanovsky 
Link: https://lore.kernel.org/r/20191212091214.315005-3-leon@kernel.org
Signed-off-by: Doug Ledford

RDMA/counter: Prevent auto-binding a QP which are not tracked with res

2019-12-12T20:38:15+00:00

Some QPs (e.g. XRC QP) are not tracked in kernel, in this case they have
an invalid res and should not be bound to any dynamically-allocated
counter in auto mode.

This fixes below call trace:
BUG: kernel NULL pointer dereference, address: 0000000000000390
PGD 80000001a7233067 P4D 80000001a7233067 PUD 1a7215067 PMD 0
Oops: 0000 [#1] SMP PTI
CPU: 2 PID: 24822 Comm: ibv_xsrq_pingpo Not tainted 5.4.0-rc5+ #21
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-2.fc27 04/01/2014
RIP: 0010:rdma_counter_bind_qp_auto+0x142/0x270 [ib_core]
Code: e1 48 85 c0 48 89 c2 0f 84 bc 00 00 00 49 8b 06 48 39 42 48 75 d6 40 3a aa 90 00 00 00 75 cd 49 8b 86 00 01 00 00 48 8b 4a 28 <8b> 80 90 03 00 00 39 81 90 03 00 00 75 b4 85 c0 74 b0 48 8b 04 24
RSP: 0018:ffffc900003f39c0 EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000000000
RDX: ffff88820020ec00 RSI: 0000000000000004 RDI: ffffffffffffffc0
RBP: 0000000000000001 R08: ffff888224149ff0 R09: ffffc900003f3968
R10: ffffffffffffffff R11: ffff8882249c5848 R12: ffffffffffffffff
R13: ffff88821d5aca50 R14: ffff8881f7690800 R15: ffff8881ff890000
FS:  00007fe53a3e1740(0000) GS:ffff888237b00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000390 CR3: 00000001a7292006 CR4: 00000000003606a0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 _ib_modify_qp+0x3a4/0x3f0 [ib_core]
 ? lookup_get_idr_uobject.part.8+0x23/0x40 [ib_uverbs]
 modify_qp+0x322/0x3e0 [ib_uverbs]
 ib_uverbs_modify_qp+0x43/0x70 [ib_uverbs]
 ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0xb1/0xf0 [ib_uverbs]
 ib_uverbs_run_method+0x6be/0x760 [ib_uverbs]
 ? uverbs_disassociate_api+0xd0/0xd0 [ib_uverbs]
 ib_uverbs_cmd_verbs+0x18d/0x3a0 [ib_uverbs]
 ? get_acl+0x1a/0x120
 ? __alloc_pages_nodemask+0x15d/0x2c0
 ib_uverbs_ioctl+0xa7/0x110 [ib_uverbs]
 do_vfs_ioctl+0xa5/0x610
 ksys_ioctl+0x60/0x90
 __x64_sys_ioctl+0x16/0x20
 do_syscall_64+0x48/0x110
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: 99fa331dc862 ("RDMA/counter: Add "auto" configuration mode support")
Signed-off-by: Mark Zhang 
Reviewed-by: Maor Gottlieb 
Reviewed-by: Ido Kalir 
Signed-off-by: Leon Romanovsky 
Link: https://lore.kernel.org/r/20191212091214.315005-2-leon@kernel.org
Signed-off-by: Doug Ledford

rxe: correctly calculate iCRC for unaligned payloads

2019-12-09T18:55:26+00:00

If RoCE PDUs being sent or received contain pad bytes, then the iCRC
is miscalculated, resulting in PDUs being emitted by RXE with an incorrect
iCRC, as well as ingress PDUs being dropped due to erroneously detecting
a bad iCRC in the PDU.  The fix is to include the pad bytes, if any,
in iCRC computations.

Note: This bug has caused broken on-the-wire compatibility with actual
hardware RoCE devices since the soft-RoCE driver was first put into the
mainstream kernel.  Fixing it will create an incompatibility with the
original soft-RoCE devices, but is necessary to be compatible with real
hardware devices.

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Steve Wise 
Link: https://lore.kernel.org/r/20191203020319.15036-2-larrystevenwise@gmail.com
Signed-off-by: Doug Ledford

treewide: Use sizeof_field() macro

2019-12-09T18:36:44+00:00

Replace all the occurrences of FIELD_SIZEOF() with sizeof_field() except
at places where these are defined. Later patches will remove the unused
definition of FIELD_SIZEOF().

This patch is generated using following script:

EXCLUDE_FILES="include/linux/stddef.h|include/linux/kernel.h"

git grep -l -e "\bFIELD_SIZEOF\b" | while read file;
do

	if [[ "$file" =~ $EXCLUDE_FILES ]]; then
		continue
	fi
	sed -i  -e 's/\bFIELD_SIZEOF\b/sizeof_field/g' $file;
done

Signed-off-by: Pankaj Bharadiya 
Link: https://lore.kernel.org/r/20190924105839.110713-3-pankaj.laxminarayan.bharadiya@intel.com
Co-developed-by: Kees Cook 
Signed-off-by: Kees Cook 
Acked-by: David Miller  # for net

RDMA/cma: add missed unregister_pernet_subsys in init failure

2019-12-09T17:02:11+00:00

The driver forgets to call unregister_pernet_subsys() in the error path
of cma_init().
Add the missed call to fix it.

Fixes: 4be74b42a6d0 ("IB/cma: Separate port allocation to network namespaces")
Signed-off-by: Chuhong Yuan 
Reviewed-by: Parav Pandit 
Link: https://lore.kernel.org/r/20191206012426.12744-1-hslester96@gmail.com
Signed-off-by: Doug Ledford

net: ipv6_stub: use ip6_dst_lookup_flow instead of ip6_dst_lookup

2019-12-04T20:27:13+00:00

ipv6_stub uses the ip6_dst_lookup function to allow other modules to
perform IPv6 lookups. However, this function skips the XFRM layer
entirely.

All users of ipv6_stub->ip6_dst_lookup use ip_route_output_flow (via the
ip_route_output_key and ip_route_output helpers) for their IPv4 lookups,
which calls xfrm_lookup_route(). This patch fixes this inconsistent
behavior by switching the stub to ip6_dst_lookup_flow, which also calls
xfrm_lookup_route().

This requires some changes in all the callers, as these two functions
take different arguments and have different return types.

Fixes: 5f81bd2e5d80 ("ipv6: export a stub for IPv6 symbols used by vxlan")
Reported-by: Xiumei Mu 
Signed-off-by: Sabrina Dubroca 
Signed-off-by: David S. Miller