platform/kernel/linux-rpi3.git
6 years agoIB/Hfi1: Read CCE Revision register to verify the device is responsive
Kamenee Arumugam [Wed, 2 May 2018 13:43:31 +0000 (06:43 -0700)]
IB/Hfi1: Read CCE Revision register to verify the device is responsive

When Hfi1 device is unresponsive, reading the RcvArrayCnt register
will return all 1's. This value is then used to remap chip's RcvArray.
The incorrect all ones value used in remapping RcvArray
will cause warn on as shown by trace below:

[<ffffffff81685eac>] dump_stack+0x19/0x1b
[<ffffffff81085820>] warn_slowpath_common+0x70/0xb0
[<ffffffff810858bc>] warn_slowpath_fmt+0x5c/0x80
[<ffffffff81065c29>] __ioremap_caller+0x279/0x320
[<ffffffff8142873c>] ? _dev_info+0x6c/0x90
[<ffffffffa021d155>] ? hfi1_pcie_ddinit+0x1d5/0x330 [hfi1]
[<ffffffff81065d62>] ioremap_wc+0x32/0x40
[<ffffffffa021d155>] hfi1_pcie_ddinit+0x1d5/0x330 [hfi1]
[<ffffffffa0204851>] hfi1_init_dd+0x1d1/0x2440 [hfi1]
[<ffffffff813503dc>] ? pci_write_config_word+0x1c/0x20

Read CCE revision register first to verify that WFR device is
responsive. If the read return "all ones", bail out from init
and fail the driver load.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Rework fault injection machinery
Mitko Haralanov [Wed, 2 May 2018 13:43:24 +0000 (06:43 -0700)]
IB/hfi1: Rework fault injection machinery

The packet fault injection code present in the HFI1 driver had some
issues which not only fragment the code but also created user
confusion. Furthermore, it suffered from the following issues:

  1. The fault_packet method only worked for received packets. This
     meant that the only fault injection mode available for sent
     packets is fault_opcode, which did not allow for random packet
     drops on all egressing packets.
  2. The mask available for the fault_opcode mode did not really work
     due to the fact that the opcode values are not bits in a bitmask but
     rather sequential integer values. Creating a opcode/mask pair that
     would successfully capture a set of packets was nearly impossible.
  3. The code was fragmented and used too many debugfs entries to
     operate and control. This was confusing to users.
  4. It did not allow filtering fault injection on a per direction basis -
     egress vs. ingress.

In order to improve or fix the above issues, the following changes have
been made:

   1. The fault injection methods have been combined into a single fault
      injection facility. As such, the fault injection has been plugged
      into both the send and receive code paths. Regardless of method used
      the fault injection will operate on both egress and ingress packets.
   2. The type of fault injection - by packet or by opcode - is now controlled
      by changing the boolean value of the file "opcode_mode". When the value
      is set to True, fault injection is done by opcode. Otherwise, by
      packet.
   2. The masking ability has been removed in favor of a bitmap that holds
      opcodes of interest (one bit per opcode, a total of 256 bits). This
      works in tandem with the "opcode_mode" value. When the value of
      "opcode_mode" is False, this bitmap is ignored. When the value is
      True, the bitmap lists all opcodes to be considered for fault injection.
      By default, the bitmap is empty. When the user wants to filter by opcode,
      the user sets the corresponding bit in the bitmap by echo'ing the bit
      position into the 'opcodes' file. This gets around the issue that the set
      of opcodes does not lend itself to effective masks and allow for extremely
      fine-grained filtering by opcode.
   4. fault_packet and fault_opcode methods have been combined. Hence, there
      is only one debugfs directory controlling the entire operation of the
      fault injection machinery. This reduces the number of debugfs entries
      and provides a more unified user experience.
   5. A new control files - "direction" - is provided to allow the user to
      control the direction of packets, which are subject to fault injection.
   6. A new control file - "skip_usec" - is added that would allow the user
      to specify a "timeout" during which no fault injection will occur.

In addition, the following bug fixes have been applied:

   1. The fault injection code has been split into its own header and source
      files. This was done to better organize the code and support conditional
      compilation without littering the code with #ifdef's.
   2. The method by which the TX PIO packets were being marked for drop
      conflicted with the way send contexts were being setup. As a result,
      the send context was repeatedly being reset.
   3. The fault injection only makes sense when the user can control it
      through the debugfs entries. However, a kernel configuration can
      enable fault injection but keep fault injection debugfs entries
      disabled. Therefore, it makes sense that the HFI fault injection
      code depends on both.
   4. Error suppression did not take into account the method by which PIO
      packets were being dropped. Therefore, even with error suppression
      turned on, errors would still be displayed to the screen. A larger
      enough packet drop percentage would case the kernel to crash because
      the driver would be stuck printing errors.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Don Hiatt <don.hiatt@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/{hfi1, qib}: Add handling of kernel restart
Alex Estrin [Wed, 2 May 2018 13:43:15 +0000 (06:43 -0700)]
IB/{hfi1, qib}: Add handling of kernel restart

A warm restart will fail to unload the driver, leaving link state
potentially flapping up to the point the BIOS resets the adapter.
Correct the issue by hooking the shutdown pci method,
which will bring port down.

Cc: <stable@vger.kernel.org> # 4.9.x
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Reorder incorrect send context disable
Michael J. Ruhl [Wed, 2 May 2018 13:43:07 +0000 (06:43 -0700)]
IB/hfi1: Reorder incorrect send context disable

User send context integrity bits are cleared before the context is
disabled.  If the send context is still processing data, any packets
that need those integrity bits will cause an error and halt the send
context.

During the disable handling, the driver waits for the context to drain.
If the context is halted, the driver will eventually timeout because
the context won't drain and then incorrectly bounce the link.

Reorder the bit clearing and the context disable.

Examine the software state and send context status as well as the
egress status to determine if a send context is in the halted state.

Promote the check macros to static functions for consistency with the
new check and to follow kernel style.

Remove an unused define that refers to the egress timeout.

Cc: <stable@vger.kernel.org> # 4.9.x
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Return correct value for device state
Michael J. Ruhl [Wed, 2 May 2018 13:42:59 +0000 (06:42 -0700)]
IB/hfi1: Return correct value for device state

The driver_pstate() function is used to map internal driver state
information to externally defined states.

The VERIFY_CAP and GOING_UP states are config/training states, but
the mapping routing returns the POLLING value.

Update the return values for VERIFY_CAP and GOING_UP to return the
correct value: TRAINING.

Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Fix fault injection init/exit issues
Mike Marciniszyn [Wed, 2 May 2018 13:42:44 +0000 (06:42 -0700)]
IB/hfi1: Fix fault injection init/exit issues

There are config dependent code paths that expose panics in unload
paths both in this file and in debugfs_remove_recursive() because
CONFIG_FAULT_INJECTION and CONFIG_FAULT_INJECTION_DEBUG_FS can be
set independently.

Having CONFIG_FAULT_INJECTION set and CONFIG_FAULT_INJECTION_DEBUG_FS
reset causes fault_create_debugfs_attr() to return an error.

The debugfs.c routines tolerate failures, but the module unload panics
dereferencing a NULL in the two exit routines.  If that is fixed, the
dir passed to debugfs_remove_recursive comes from a memory location
that was freed and potentially reused causing a segfault or corrupting
memory.

Here is an example of the NULL deref panic:

[66866.286829] BUG: unable to handle kernel NULL pointer dereference at 0000000000000088
[66866.295602] IP: hfi1_dbg_ibdev_exit+0x2a/0x80 [hfi1]
[66866.301138] PGD 858496067 P4D 858496067 PUD 8433a7067 PMD 0
[66866.307452] Oops: 0000 [#1] SMP
[66866.310953] Modules linked in: hfi1(-) rdmavt rdma_ucm ib_ucm ib_uverbs ib_umad rdma_cm iw_cm ib_cm ib_core rpcsec_gss_krb5 nfsv4 dns_resolver nfsv3 nfs fscache sb_edac x86_pkg_temp_thermal intel_powerclamp vfat fat coretemp kvm irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel iTCO_wdt iTCO_vendor_support crypto_simd mei_me glue_helper cryptd mxm_wmi ipmi_si pcspkr lpc_ich sg mei ioatdma ipmi_devintf i2c_i801 mfd_core shpchp ipmi_msghandler wmi acpi_power_meter acpi_cpufreq nfsd auth_rpcgss nfs_acl lockd grace sunrpc ip_tables ext4 mbcache jbd2 sd_mod mgag200 drm_kms_helper syscopyarea sysfillrect sysimgblt igb fb_sys_fops ttm ahci ptp crc32c_intel libahci pps_core drm dca libata i2c_algo_bit i2c_core [last unloaded: opa_vnic]
[66866.385551] CPU: 8 PID: 7470 Comm: rmmod Not tainted 4.14.0-mam-tid-rdma #2
[66866.393317] Hardware name: Intel Corporation S2600WT2/S2600WT2, BIOS SE5C610.86B.01.01.0018.C4.072020161249 07/20/2016
[66866.405252] task: ffff88084f28c380 task.stack: ffffc90008454000
[66866.411866] RIP: 0010:hfi1_dbg_ibdev_exit+0x2a/0x80 [hfi1]
[66866.417984] RSP: 0018:ffffc90008457da0 EFLAGS: 00010202
[66866.423812] RAX: 0000000000000000 RBX: ffff880857de0000 RCX: 0000000180040001
[66866.431773] RDX: 0000000180040002 RSI: ffffea0021088200 RDI: 0000000040000000
[66866.439734] RBP: ffffc90008457da8 R08: ffff88084220e000 R09: 0000000180040001
[66866.447696] R10: 000000004220e001 R11: ffff88084220e000 R12: ffff88085a31c000
[66866.455657] R13: ffffffffa07c9820 R14: ffffffffa07c9890 R15: ffff881059d78100
[66866.463618] FS:  00007f6876047740(0000) GS:ffff88085f800000(0000) knlGS:0000000000000000
[66866.472644] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[66866.479053] CR2: 0000000000000088 CR3: 0000000856357006 CR4: 00000000001606e0
[66866.487013] Call Trace:
[66866.489747]  remove_one+0x1f/0x220 [hfi1]
[66866.494221]  pci_device_remove+0x39/0xc0
[66866.498596]  device_release_driver_internal+0x141/0x210
[66866.504424]  driver_detach+0x3f/0x80
[66866.508409]  bus_remove_driver+0x55/0xd0
[66866.512784]  driver_unregister+0x2c/0x50
[66866.517164]  pci_unregister_driver+0x2a/0xa0
[66866.521934]  hfi1_mod_cleanup+0x10/0xaa2 [hfi1]
[66866.526988]  SyS_delete_module+0x171/0x250
[66866.531558]  do_syscall_64+0x67/0x1b0
[66866.535644]  entry_SYSCALL64_slow_path+0x25/0x25
[66866.540792] RIP: 0033:0x7f6875525c27
[66866.544777] RSP: 002b:00007ffd48528e78 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0
[66866.553224] RAX: ffffffffffffffda RBX: 0000000001cc01d0 RCX: 00007f6875525c27
[66866.561185] RDX: 00007f6875596000 RSI: 0000000000000800 RDI: 0000000001cc0238
[66866.569146] RBP: 0000000000000000 R08: 00007f68757e9060 R09: 00007f6875596000
[66866.577120] R10: 00007ffd48528c00 R11: 0000000000000206 R12: 00007ffd48529db4
[66866.585080] R13: 0000000000000000 R14: 0000000001cc01d0 R15: 0000000001cc0010
[66866.593040] Code: 90 0f 1f 44 00 00 48 83 3d a3 8b 03 00 00 55 48 89 e5 53 48 89 fb 74 4e 48 8d bf 18 0c 00 00 e8 9d f2 ff ff 48 8b 83 20 0c 00 00 <48> 8b b8 88 00 00 00 e8 2a 21 b3 e0 48 8b bb 20 0c 00 00 e8 0e
[66866.614127] RIP: hfi1_dbg_ibdev_exit+0x2a/0x80 [hfi1] RSP: ffffc90008457da0
[66866.621885] CR2: 0000000000000088
[66866.625618] ---[ end trace c4817425783fb092 ]---

Fix by insuring that upon failure from fault_create_debugfs_attr() the
parent pointer for the routines is always set to NULL and guards added
in the exit routines to insure that debugfs_remove_recursive() is not
called when when the parent pointer is NULL.

Fixes: 0181ce31b260 ("IB/hfi1: Add receive fault injection feature")
Cc: <stable@vger.kernel.org> # 4.14.x
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Complete check for locally terminated smp
Alex Estrin [Wed, 2 May 2018 13:42:36 +0000 (06:42 -0700)]
IB/hfi1: Complete check for locally terminated smp

For lid routed packets 'hop_cnt' is zero, therefore current
test is incomplete. Fix it by using local mad check for
both lid routed and direct routed MADs.

Reviewed-by: Mike Mariciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Return actual error value from program_rcvarray()
Michael J. Ruhl [Wed, 2 May 2018 13:42:29 +0000 (06:42 -0700)]
IB/hfi1: Return actual error value from program_rcvarray()

A failure of program_rcvarray() is treated inconsistently by the
calling function.  In one case the error is returned, in a second
case, the error is overwritten with EFAULT.  In both cases the
code path is doing the same thing, allocating memory for groups,
so it should be consistent.

Make the error path consistent and return the error generated by
program_rcvarray().

Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Fixes: 7e7a436ecb6e ("staging/hfi1: Add TID entry program function body")
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Prevent LNI hang when LCB can't obtain lanes
Sebastian Sanchez [Wed, 2 May 2018 13:42:21 +0000 (06:42 -0700)]
IB/hfi1: Prevent LNI hang when LCB can't obtain lanes

When the LCB isn't able to get any lanes operational on the
first transition into mission mode, the link transfer active
never happens and the LNI stays in the polling state indefinitely.

Reset LCB upon receiving an 8051 interrupt for LCB to try to obtain
lanes with firmware version 1.25.0 or later. Also, update the LCB
reset value in other parts of the code with a macro defined to make
the code more maintainable and rename functions with the link_width
label to link_mode to reflect the fact that those functions set and
read link related data not just the link width.

Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoMerge branch 'k.o/for-rc' into k.o/wip/dl-for-next
Doug Ledford [Wed, 9 May 2018 19:48:48 +0000 (15:48 -0400)]
Merge branch 'k.o/for-rc' into k.o/wip/dl-for-next

Several items of conflict have arisen between the RDMA stack's for-rc
branch and upcoming for-next work:

9fd4350ba895 ("IB/rxe: avoid double kfree_skb") directly conflicts with
2e47350789eb ("IB/rxe: optimize the function duplicate_request")

Patches already submitted by Intel for the hfi1 driver will fail to
apply cleanly without this merge

Other people on the mailing list have notified that their upcoming
patches also fail to apply cleanly without this merge

Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/mlx5: posting klm/mtt list inline in the send queue for reg_wr
Idan Burstein [Wed, 2 May 2018 10:16:39 +0000 (13:16 +0300)]
IB/mlx5: posting klm/mtt list inline in the send queue for reg_wr

As most kernel RDMA ULPs, (e.g. NVMe over Fabrics in its default
"register_always=Y" mode) registers and invalidates user buffer
upon each IO.

Today the mlx5 driver is posting the registration work
request using scatter/gather entry for the MTT/KLM list.
The fetch of the MTT/KLM list becomes the bottleneck in
number of IO operation could be done by NVMe over Fabrics
host driver on a single adapter as shown below.

This patch is adding the support for inline registration
work request upon MTT/KLM list of size <=64B.

The result for NVMe over Fabrics is increase of > x3.5 for small
IOs as shown below, I expect other ULPs (e.g iSER, SRP, NFS over RDMA)
performance to be enhanced as well.

The following results were taken against a single NVMe-oF (RoCE link layer)
subsystem with a single namespace backed by null_blk using fio benchmark
(with rw=randread, numjobs=48, iodepth={16,64}, ioengine=libaio direct=1):

ConnectX-5 (pci Width x16)
---------------------------

Block Size       s/g reg_wr            inline reg_wr
++++++++++     +++++++++++++++        ++++++++++++++++
512B            1302.8K/34.82%         4951.9K/99.02%
1KB             1284.3K/33.86%         4232.7K/98.09%
2KB             1238.6K/34.1%          2797.5K/80.04%
4KB             1169.3K/32.46%         1941.3K/61.35%
8KB             1013.4K/30.08%         1236.6K/39.47%
16KB            695.7K/20.19%          696.9K/20.59%
32KB            350.3K/9.64%           350.6K/10.3%
64KB            175.86K/5.27%          175.9K/5.28%

ConnectX-4 (pci Width x8)
---------------------------

Block Size       s/g reg_wr            inline reg_wr
++++++++++     +++++++++++++++        ++++++++++++++++
512B            1285.8K/42.66%          4242.7K/98.18%
1KB             1254.1K/41.74%          3569.2K/96.00%
2KB             1185.9K/39.83%          2173.9K/75.58%
4KB             1069.4K/36.46%          1343.3K/47.47%
8KB             755.1K/27.77%           748.7K/29.14%

Tested-by: Nitzan Carmi <nitzanc@mellanox.com>
Signed-off-by: Idan Burstein <idanb@mellanox.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Drop local zgid in favor of core defined variable
Leon Romanovsky [Wed, 2 May 2018 10:15:24 +0000 (13:15 +0300)]
RDMA/hns: Drop local zgid in favor of core defined variable

The zgid is already provided by IB/core, so there is no need in locally
defined variable, let's drop it and reuse common one.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Reuse gid_table_release_one() in table allocation failure
Parav Pandit [Wed, 2 May 2018 10:12:56 +0000 (13:12 +0300)]
IB/core: Reuse gid_table_release_one() in table allocation failure

_gid_table_setup_one() only performs GID table cache memory allocation,
marks entries as invalid (free) and marks the reserved entries.
At this point GID table is empty and no entries are added.

On dual port device if _gid_table_setup_one() fails to allocate the gid
table for 2nd port, there is no need to perform cleanup_gid_table_port()
to delete GID entries, as GID table is empty.
Therefore make use of existing gid_table_release_one() routine which
frees the GID table memory and avoid code duplication.

Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Make gid_table_reserve_default() return void
Parav Pandit [Wed, 2 May 2018 10:12:55 +0000 (13:12 +0300)]
IB/core: Make gid_table_reserve_default() return void

gid_table_reserve_default() always returns zero. Make it return void and
simplify error checking.

rdma_port is already calculated, use that while calling
gid_table_reserve_default() instead of recalculating it.

Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoiw_cxgb4: dump detailed driver-specific QP information
Steve Wise [Thu, 3 May 2018 15:41:49 +0000 (08:41 -0700)]
iw_cxgb4: dump detailed driver-specific QP information

Provide a cxgb4-specific function to fill in qp state details.
This allows dumping important c4iw_qp state useful for debugging.

Included in the dump are the t4_sq, t4_rq structs, plus a dump
of the t4_swsqe and t4swrqe descriptors for the first and last
pending entries.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/nldev: helper functions to add driver attributes
Steve Wise [Thu, 3 May 2018 15:41:42 +0000 (08:41 -0700)]
RDMA/nldev: helper functions to add driver attributes

These help rdma drivers to fill out the driver entries.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/nldev: add driver-specific resource tracking
Steve Wise [Thu, 3 May 2018 15:41:30 +0000 (08:41 -0700)]
RDMA/nldev: add driver-specific resource tracking

Each driver can register a "fill entry" function with the restrack core.
This function will be called when filling out a resource, allowing the
driver to add driver-specific details.  The details consist of a
nltable of nested attributes, that are in the form of <key, [print-type],
value> tuples.  Both key and value attributes are mandatory.  The key
nlattr must be a string, and the value nlattr can be one of the driver
attributes that are generic, but typed, allowing the attributes to be
validated.  Currently the driver nlattr types include string, s32,
u32, s64, and u64.  The print-type nlattr allows a driver to specify
an alternative display format for user tools displaying the attribute.
For example, a u32 attribute will default to "%u", but a print-type
attribute can be included for it to be displayed in hex.  This allows
the user tool to print the number in the format desired by the driver
driver.

More attrs can be defined as they become needed by drivers.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/nldev: Add explicit pad attribute
Steve Wise [Thu, 3 May 2018 15:40:49 +0000 (08:40 -0700)]
RDMA/nldev: Add explicit pad attribute

Add a specific RDMA_NLDEV_ATTR_PAD attribute to be used for 64b
attribute padding.  To preserve the ABI, make this attribute equal to
RDMA_NLDEV_ATTR_UNSPEC, which has a value of 0, because that has been
used up until now as the pad attribute.

Change all the previous use of 0 as the pad with this
new enum.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/cma: Do not query GID during QP state transition to RTR
Parav Pandit [Wed, 2 May 2018 10:18:59 +0000 (13:18 +0300)]
RDMA/cma: Do not query GID during QP state transition to RTR

When commit [1] was added, SGID was queried to derive the SMAC address.
Then, later on during a refactor [2], SMAC was no longer needed. However,
the now useless GID query remained.  Then during additional code changes
later on, the GID query was being done in such a way that it caused iWARP
queries to start breaking.  Remove the useless GID query and resolve the
iWARP breakage at the same time.

This is discussed in [3].

[1] commit dd5f03beb4f7 ("IB/core: Ethernet L2 attributes in verbs/cm structures")
[2] commit 5c266b2304fb ("IB/cm: Remove the usage of smac and vid of qp_attr and cm_av")
[3] https://www.spinics.net/lists/linux-rdma/msg63951.html

Suggested-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/mlx4: Fix integer overflow when calculating optimal MTT size
Jack Morgenstein [Wed, 2 May 2018 10:04:25 +0000 (13:04 +0300)]
IB/mlx4: Fix integer overflow when calculating optimal MTT size

When the kernel was compiled using the UBSAN option,
we saw the following stack trace:

[ 1184.827917] UBSAN: Undefined behaviour in drivers/infiniband/hw/mlx4/mr.c:349:27
[ 1184.828114] signed integer overflow:
[ 1184.828247] -2147483648 - 1 cannot be represented in type 'int'

The problem was caused by calling round_up in procedure
mlx4_ib_umem_calc_optimal_mtt_size (on line 349, as noted in the stack
trace) with the second parameter (1 << block_shift) (which is an int).
The second parameter should have been (1ULL << block_shift) (which
is an unsigned long long).

(1 << block_shift) is treated by the compiler as an int (because 1 is
an integer).

Now, local variable block_shift is initialized to 31.
If block_shift is 31, 1 << block_shift is 1 << 31 = 0x80000000=-214748368.
This is the most negative int value.

Inside the round_up macro, there is a cast applied to ((1 << 31) - 1).
However, this cast is applied AFTER ((1 << 31) - 1) is calculated.
Since (1 << 31) is treated as an int, we get the negative overflow
identified by UBSAN in the process of calculating ((1 << 31) - 1).

The fix is to change (1 << block_shift) to (1ULL << block_shift) on
line 349.

Fixes: 9901abf58368 ("IB/mlx4: Use optimal numbers of MTT entries")
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Fix memory leak in exception path in get_irq_affinity()
Sebastian Sanchez [Tue, 1 May 2018 12:36:13 +0000 (05:36 -0700)]
IB/hfi1: Fix memory leak in exception path in get_irq_affinity()

When IRQ affinity is set and the interrupt type is unknown, a cpu
mask allocated within the function is never freed. Fix this memory
leak by allocating memory within the scope where it is used.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/{hfi1, rdmavt}: Fix memory leak in hfi1_alloc_devdata() upon failure
Sebastian Sanchez [Tue, 1 May 2018 12:36:06 +0000 (05:36 -0700)]
IB/{hfi1, rdmavt}: Fix memory leak in hfi1_alloc_devdata() upon failure

When allocating device data, if there's an allocation failure, the
already allocated memory won't be freed such as per-cpu counters.

Fix memory leaks in exception path by creating a common reentrant
clean up function hfi1_clean_devdata() to be used at driver unload
time and device data allocation failure.

To accomplish this, free_platform_config() and clean_up_i2c() are
changed to be reentrant to remove dependencies when they are called
in different order. This helps avoid NULL pointer dereferences
introduced by this patch if those two functions weren't reentrant.

In addition, set dd->int_counter, dd->rcv_limit,
dd->send_schedule and dd->tx_opstats to NULL after they're freed in
hfi1_clean_devdata(), so that hfi1_clean_devdata() is fully reentrant.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Fix NULL pointer dereference when invalid num_vls is used
Sebastian Sanchez [Tue, 1 May 2018 12:35:58 +0000 (05:35 -0700)]
IB/hfi1: Fix NULL pointer dereference when invalid num_vls is used

When an invalid num_vls is used as a module parameter, the code
execution follows an exception path where the macro dd_dev_err()
expects dd->pcidev->dev not to be NULL in hfi1_init_dd(). This
causes a NULL pointer dereference.

Fix hfi1_init_dd() by initializing dd->pcidev and dd->pcidev->dev
earlier in the code. If a dd exists, then dd->pcidev and
dd->pcidev->dev always exists.

BUG: unable to handle kernel NULL pointer dereference
at 00000000000000f0
IP: __dev_printk+0x15/0x90
Workqueue: events work_for_cpu_fn
RIP: 0010:__dev_printk+0x15/0x90
Call Trace:
 dev_err+0x6c/0x90
 ? hfi1_init_pportdata+0x38d/0x3f0 [hfi1]
 hfi1_init_dd+0xdd/0x2530 [hfi1]
 ? pci_conf1_read+0xb2/0xf0
 ? pci_read_config_word.part.9+0x64/0x80
 ? pci_conf1_write+0xb0/0xf0
 ? pcie_capability_clear_and_set_word+0x57/0x80
 init_one+0x141/0x490 [hfi1]
 local_pci_probe+0x3f/0xa0
 work_for_cpu_fn+0x10/0x20
 process_one_work+0x152/0x350
 worker_thread+0x1cf/0x3e0
 kthread+0xf5/0x130
 ? max_active_store+0x80/0x80
 ? kthread_bind+0x10/0x10
 ? do_syscall_64+0x6e/0x1a0
 ? SyS_exit_group+0x10/0x10
 ret_from_fork+0x35/0x40

Cc: <stable@vger.kernel.org> # 4.9.x
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Fix loss of BECN with AHG
Mike Marciniszyn [Tue, 1 May 2018 12:35:51 +0000 (05:35 -0700)]
IB/hfi1: Fix loss of BECN with AHG

AHG may be armed to use the stored header, which by design is limited
to edits in the PSN/A 32 bit word (bth2).

When the code is trying to send a BECN, the use of the stored header
will lose the BECN bit.

Fix by avoiding AHG when getting ready to send a BECN. This is
accomplished by always claiming the packet is not a middle packet which
is an AHG precursor.  BECNs are not a normal case and this should not
hurt AHG optimizations.

Cc: <stable@vger.kernel.org> # 4.14.x
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1 Use correct type for num_user_context
Michael J. Ruhl [Tue, 1 May 2018 12:35:43 +0000 (05:35 -0700)]
IB/hfi1 Use correct type for num_user_context

The module parameter num_user_context is defined as 'int' and
defaults to -1.  The module_param_named() says that it is uint.

Correct module_param_named() type information and update the modinfo
text to reflect the default value.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/hfi1: Fix handling of FECN marked multicast packet
Mike Marciniszyn [Tue, 1 May 2018 12:35:36 +0000 (05:35 -0700)]
IB/hfi1: Fix handling of FECN marked multicast packet

The code for handling a marked UD packet unconditionally returns the
dlid in the header of the FECN marked packet.  This is not correct
for multicast packets where the DLID is in the multicast range.

The subsequent attempt to send the CNP with the multicast lid will
cause the chip to halt the ack send context because the source
lid doesn't match the chip programming.   The send context will
be halted and flush any other pending packets in the pio ring causing
the CNP to not be sent.

A part of investigating the fix, it was determined that the 16B work
broke the FECN routine badly with inconsistent use of 16 bit and 32 bits
types for lids and pkeys.  Since the port's source lid was correctly 32
bits the type mixmatches need to be dealt with at the same time as
fixing the CNP header issue.

Fix these issues by:
- Using the ports lid for as the SLID for responding to FECN marked UD
  packets
- Insure pkey is always 16 bit in this and subordinate routines
- Insure lids are 32 bits in this and subordinate routines

Cc: <stable@vger.kernel.org> # 4.14.x
Fixes: 88733e3b8450 ("IB/hfi1: Add 16B UD support")
Reviewed-by: Don Hiatt <don.hiatt@intel.com>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/qedr: fix spelling mistake: "failes" -> "fails"
Colin Ian King [Tue, 1 May 2018 08:25:49 +0000 (09:25 +0100)]
RDMA/qedr: fix spelling mistake: "failes" -> "fails"

Trivial fix to spelling mistake in DP_ERR error message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/cxgb4: use skb_put_zero()/__skb_put_zero
YueHaibing [Sat, 28 Apr 2018 07:31:06 +0000 (15:31 +0800)]
IB/cxgb4: use skb_put_zero()/__skb_put_zero

Use the recently introduced helper to replace the pattern of
skb_put_zero/__skb_put() && memset().

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Use CONFIG_SECURITY_INFINIBAND to compile out security code
Parav Pandit [Thu, 26 Apr 2018 07:56:34 +0000 (10:56 +0300)]
IB/core: Use CONFIG_SECURITY_INFINIBAND to compile out security code

Make security.c depends on CONFIG_SECURITY_INFINIBAND.

Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Make ib_mad_client_id atomic
HÃ¥kon Bugge [Wed, 18 Apr 2018 14:24:50 +0000 (16:24 +0200)]
IB/core: Make ib_mad_client_id atomic

Currently, the kernel protects access to the agent ID allocator on a per
port basis using a spinlock, so it is impossible for two apps/threads on
the same port to get the same TID, but it is entirely possible for two
threads on different ports to end up with the same TID.

As this can be confusing (regardless of it being legal according to the
IB Spec 1.3, C13-18.1.1, in section 13.4.6.4 - TransactionID usage),
and as the rdma-core user space API for /dev/umad devices implies unique
TIDs even across ports, make the TID an atomic type so that no two
allocations, regardless of port number, will be the same.

Signed-off-by: HÃ¥kon Bugge <haakon.bugge@oracle.com>
Reviewed-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoiw_cxgb4: Atomically flush per QP HW CQEs
Bharat Potnuri [Fri, 27 Apr 2018 11:11:16 +0000 (16:41 +0530)]
iw_cxgb4: Atomically flush per QP HW CQEs

When a CQ is shared by multiple QPs, c4iw_flush_hw_cq() needs to acquire
corresponding QP lock before moving the CQEs into its corresponding SW
queue and accessing the SQ contents for completing a WR.
Ignore CQEs if corresponding QP is already flushed.

Cc: stable@vger.kernel.org
Signed-off-by: Potnuri Bharat Teja <bharat@chelsio.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/uverbs: Fix kernel crash during MR deregistration flow
Ariel Levkovich [Thu, 26 Apr 2018 12:42:55 +0000 (15:42 +0300)]
IB/uverbs: Fix kernel crash during MR deregistration flow

This patch fixes a crash that happens due to access to an
uninitialized DM pointer within the MR object.

The change makes sure the DM pointer in the MR object is set to
NULL during a non-DM MR creation to prevent a false indication
that this MR is related to a DM in the dereg flow.

Fixes: be934cca9e98 ("IB/uverbs: Add device memory registration ioctl support")
Reported-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Ariel Levkovich <lariel@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/uverbs: Prevent reregistration of DM_MR to regular MR
Ariel Levkovich [Thu, 26 Apr 2018 12:42:54 +0000 (15:42 +0300)]
IB/uverbs: Prevent reregistration of DM_MR to regular MR

This patch adds a check in the ib_uverbs_rereg_mr flow to make
sure there's no attempt to rereg a device memory MR to regular MR.
In such case the command will fail with -EINVAL status.

fixes: be934cca9e98 ("IB/uverbs: Add device memory registration ioctl support")
Signed-off-by: Ariel Levkovich <lariel@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/mlx4: Add missed RSS hash inner header flag
Leon Romanovsky [Thu, 26 Apr 2018 12:37:48 +0000 (15:37 +0300)]
RDMA/mlx4: Add missed RSS hash inner header flag

Despite being advertised to user space application, the RSS inner
header flag was filtered by checks at the beginning of QP creation
routine.

Cc: <stable@vger.kernel.org> # 4.15
Fixes: 4d02ebd9bbbd ("IB/mlx4: Fix RSS hash fields restrictions")
Fixes: 07d84f7b6adf ("IB/mlx4: Add support to RSS hash for inner headers")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Fix a couple misspellings
oulijun [Thu, 26 Apr 2018 06:46:24 +0000 (14:46 +0800)]
RDMA/hns: Fix a couple misspellings

This patch fixes two spelling errors.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Submit bad wr
oulijun [Thu, 26 Apr 2018 06:46:23 +0000 (14:46 +0800)]
RDMA/hns: Submit bad wr

When generated bad work reqeust, it needs to
report to user. This patch mainly fixes it.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Update assignment method for owner field of send wqe
oulijun [Thu, 26 Apr 2018 06:46:22 +0000 (14:46 +0800)]
RDMA/hns: Update assignment method for owner field of send wqe

When posting a work reqeust, it need to update the owner bit of send
wqe. This patch mainly fix the bug when posting multiply work
request.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Adjust the order of cleanup hem table
oulijun [Thu, 26 Apr 2018 06:46:21 +0000 (14:46 +0800)]
RDMA/hns: Adjust the order of cleanup hem table

This patch update the order of cleaning hem table for trrl_table and irrl_table
as well as mtt_cqe_table and mtt_table.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Only assign dqpn if IB_QP_PATH_DEST_QPN bit is set
oulijun [Thu, 26 Apr 2018 06:46:20 +0000 (14:46 +0800)]
RDMA/hns: Only assign dqpn if IB_QP_PATH_DEST_QPN bit is set

Only when the IB_QP_PATH_DEST_QPN flag of attr_mask is set
is it valid to assign the dqpn field of qp context

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Remove some unnecessary attr_mask judgement
oulijun [Thu, 26 Apr 2018 06:46:19 +0000 (14:46 +0800)]
RDMA/hns: Remove some unnecessary attr_mask judgement

This patch deletes some unnecessary attr_mask if condition
in hip08 according to the IB protocol.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Only assign mtu if IB_QP_PATH_MTU bit is set
oulijun [Thu, 26 Apr 2018 06:46:18 +0000 (14:46 +0800)]
RDMA/hns: Only assign mtu if IB_QP_PATH_MTU bit is set

Only when the IB_QP_PATH_MTU flag of attr_mask is set
it is valid to assign the mtu field of qp context when
qp type is not GSI and UD.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Fix the qp context state diagram
oulijun [Thu, 26 Apr 2018 06:46:17 +0000 (14:46 +0800)]
RDMA/hns: Fix the qp context state diagram

According to RoCE protocol, it is possible to
transition from error to error state for modifying
qp in hip08. This patch fix it.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Intercept illegal RDMA operation when use inline data
oulijun [Thu, 26 Apr 2018 06:46:16 +0000 (14:46 +0800)]
RDMA/hns: Intercept illegal RDMA operation when use inline data

RDMA read operation is not supported inline data. If user cofigures
issue a RDMA read and use inline data, it will happen a hardware
error.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/hns: Bugfix for init hem table
oulijun [Thu, 26 Apr 2018 06:46:15 +0000 (14:46 +0800)]
RDMA/hns: Bugfix for init hem table

During init hem table, type should be used instead of
table->type which is finally initializaed with type.

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: avoid double kfree_skb
Zhu Yanjun [Thu, 26 Apr 2018 04:41:10 +0000 (00:41 -0400)]
IB/rxe: avoid double kfree_skb

When skb is sent, it will pass the following functions in soft roce.

rxe_send [rdma_rxe]
    ip_local_out
        __ip_local_out
        ip_output
            ip_finish_output
                ip_finish_output2
                    dev_queue_xmit
                        __dev_queue_xmit
                            dev_hard_start_xmit

In the above functions, if error occurs in the above functions or
iptables rules drop skb after ip_local_out, kfree_skb will be called.
So it is not necessary to call kfree_skb in soft roce module again.
Or else crash will occur.

The steps to reproduce:

     server                       client
    ---------                    ---------
    |1.1.1.1|<----rxe-channel--->|1.1.1.2|
    ---------                    ---------

On server: rping -s -a 1.1.1.1 -v -C 10000 -S 512
On client: rping -c -a 1.1.1.1 -v -C 10000 -S 512

The kernel configs CONFIG_DEBUG_KMEMLEAK and
CONFIG_DEBUG_OBJECTS are enabled on both server and client.

When rping runs, run the following command in server:

iptables -I OUTPUT -p udp  --dport 4791 -j DROP

Without this patch, crash will occur.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: add RXE_START_MASK for rxe_opcode IB_OPCODE_RC_SEND_ONLY_INV
Jianchao Wang [Thu, 26 Apr 2018 03:52:39 +0000 (11:52 +0800)]
IB/rxe: add RXE_START_MASK for rxe_opcode IB_OPCODE_RC_SEND_ONLY_INV

w/o RXE_START_MASK, the last_psn of IB_OPCODE_RC_SEND_ONLY_INV
will not be updated in update_wqe_psn, and the corresponding
wqe will not be acked in rxe_completer due to its last_psn is
zero. Finally, the other wqe will also not be able to be acked,
because the wqe of IB_OPCODE_RC_SEND_ONLY_INV with last_psn 0
is still there. This causes large amount of io timeout when
nvmeof is over rxe.

Add RXE_START_MASK for IB_OPCODE_RC_SEND_ONLY_INV to fix this.

Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Reviewed-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/iwpm: fix memory leak on map_info
Colin Ian King [Wed, 25 Apr 2018 16:24:04 +0000 (17:24 +0100)]
RDMA/iwpm: fix memory leak on map_info

In the cases where iwpm_hash_bucket is NULL and where function
get_mapinfo_hash_bucket returns NULL then the map_info is never added
to hash_bucket_head and hence there is a leak of map_info. Fix this
by nullifying hash_bucket_head and if that is null we know that
that map_info was not added to hash_bucket_head and hence map_info
should be free'd.

Detected by CoverityScan, CID#1222481 ("Resource Leak")

Fixes: 30dc5e63d6a5 ("RDMA/core: Add support for iWARP Port Mapper user space service")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/ipoib: fix ipoib_start_xmit()'s return type
Luc Van Oostenryck [Tue, 24 Apr 2018 13:15:47 +0000 (15:15 +0200)]
IB/ipoib: fix ipoib_start_xmit()'s return type

The method ndo_start_xmit() is defined as returning an 'netdev_tx_t',
which is a typedef for an enum type, but the implementation in this
driver returns an 'int'.

Fix this by returning 'netdev_tx_t' in this driver too.

Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/nes: fix nes_netdev_start_xmit()'s return type
Luc Van Oostenryck [Tue, 24 Apr 2018 13:15:45 +0000 (15:15 +0200)]
IB/nes: fix nes_netdev_start_xmit()'s return type

The method ndo_start_xmit() is defined as returning an 'netdev_tx_t',
which is a typedef for an enum type, but the implementation in this
driver returns an 'int'.

Fix this by returning 'netdev_tx_t' in this driver too.

Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/cma: Fix use after destroy access to net namespace for IPoIB
Parav Pandit [Tue, 24 Apr 2018 17:13:45 +0000 (20:13 +0300)]
RDMA/cma: Fix use after destroy access to net namespace for IPoIB

There are few issues with validation of netdevice and listen id lookup
for IB (IPoIB) while processing incoming CM request as below.

1. While performing lookup of bind_list in cma_ps_find(), net namespace
of the netdevice can get deleted in cma_exit_net(), resulting in use
after free access of idr and/or net namespace structures.
This lookup occurs from the workqueue context (and not userspace
context where net namespace is always valid).

           CPU0                              CPU1
           ====                              ====

 bind_list = cma_ps_find();
                                     move netdevice to new namespace
                                     delete net namespace
                                        cma_exit_net()
                                           idr_destroy(idr);

 [..]
 cma_find_listener(bind_list, ..);

2. While netdevice is validated for IP address in given net namespace,
netdevice's net namespace and/or ifindex can change in
cma_get_net_dev() and cma_match_net_dev().

Above issues are overcome by using rcu lock along with netdevice
UP/DOWN state as described below.
When a net namespace is getting deleted, netdevice is closed and
shutdown before moving it back to init_net namespace.
change_net_namespace() synchronizes with any existing use of netdevice
before changing the netdev properties such as net or ifindex.
Once netdevice IFF_UP flags is cleared, such fields are not guaranteed
to be valid.
Therefore, rcu lock along with netdevice state check ensures that,
while route lookup and cm_id lookup is in progress, netdevice of
interest won't migrate to any other net namespace.
This ensures that associated net namespace of netdevice won't get
deleted while rcu lock is held for netdevice which is in IFF_UP state.

Fixes: fa20105e09e9 ("IB/cma: Add support for network namespaces")
Fixes: 4be74b42a6d0 ("IB/cma: Separate port allocation to network namespaces")
Fixes: f887f2ac87c2 ("IB/cma: Validate routing of incoming requests")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/uverbs: Fix validating mandatory attributes
Matan Barak [Tue, 24 Apr 2018 08:15:20 +0000 (08:15 +0000)]
IB/uverbs: Fix validating mandatory attributes

Previously, if a method contained mandatory attributes in a namespace
that wasn't given by the user, these attributes weren't validated.
Fixing this by iterating over all specification namespaces.

Fixes: fac9658cabb9 ("IB/core: Add new ioctl interface")
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/cxgb4: release hw resources on device removal
Raju Rangoju [Mon, 23 Apr 2018 16:12:37 +0000 (21:42 +0530)]
RDMA/cxgb4: release hw resources on device removal

The c4iw_rdev_close() logic was not releasing all the hw
resources (PBL and RQT memory) during the device removal
event (driver unload / system reboot). This can cause panic
in gen_pool_destroy().

The module remove function will wait for all the hw
resources to be released during the device removal event.

Fixes c12a67fe(iw_cxgb4: free EQ queue memory on last deref)
Signed-off-by: Raju Rangoju <rajur@chelsio.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Cc: stable@vger.kernel.org
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: remove unused function variable
Zhu Yanjun [Mon, 23 Apr 2018 07:57:58 +0000 (03:57 -0400)]
IB/rxe: remove unused function variable

In the functions rxe_mem_init_dma, rxe_mem_init_user, rxe_mem_init_fast
and copy_data, the function variable rxe is not used. So this function
variable rxe is removed.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: change rxe_set_mtu function type to void
Zhu Yanjun [Fri, 20 Apr 2018 14:30:54 +0000 (10:30 -0400)]
IB/rxe: change rxe_set_mtu function type to void

The function rxe_set_mtu always returns zero. So this function type
is changed to void.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: Change rxe_rcv to return void
Yuval Shaia [Fri, 20 Apr 2018 14:05:03 +0000 (17:05 +0300)]
IB/rxe: Change rxe_rcv to return void

It always returns 0. Change return type to void.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Reviewed-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoinfiniband: hw: qib: Change return type to vm_fault_t
Souptick Joarder [Tue, 17 Apr 2018 14:34:28 +0000 (20:04 +0530)]
infiniband: hw: qib: Change return type to vm_fault_t

Use new return type vm_fault_t for fault handler. For
now, this is just documenting that the function returns
a VM_FAULT value rather than an errno. Once all instances
are converted, vm_fault_t will become a distinct type.

Reference id -> 1c8f422059ae ("mm: change return type to
vm_fault_t")

Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoinfiniband: hw: hfi1: Change return type to vm_fault_t
Souptick Joarder [Tue, 17 Apr 2018 14:23:58 +0000 (19:53 +0530)]
infiniband: hw: hfi1: Change return type to vm_fault_t

Use new return type vm_fault_t for fault handler. For
now, this is just documenting that the function returns
a VM_FAULT value rather than an errno. Once all instances
are converted, vm_fault_t will become a distinct type.

Reference id -> 1c8f422059ae ("mm: change return type to
vm_fault_t")

Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB: make INFINIBAND_ADDR_TRANS configurable
Greg Thelen [Thu, 26 Apr 2018 18:19:35 +0000 (11:19 -0700)]
IB: make INFINIBAND_ADDR_TRANS configurable

Allow INFINIBAND without INFINIBAND_ADDR_TRANS because fuzzing has been
finding fair number of CM bugs.  So provide option to disable it.

Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoib_srp: depend on INFINIBAND_ADDR_TRANS
Greg Thelen [Thu, 26 Apr 2018 18:19:34 +0000 (11:19 -0700)]
ib_srp: depend on INFINIBAND_ADDR_TRANS

INFINIBAND_SRP code depends on INFINIBAND_ADDR_TRANS provided symbols.
So declare the kconfig dependency.  This is necessary to allow for
enabling INFINIBAND without INFINIBAND_ADDR_TRANS.

Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agocifs: smbd: depend on INFINIBAND_ADDR_TRANS
Greg Thelen [Thu, 26 Apr 2018 18:19:33 +0000 (11:19 -0700)]
cifs: smbd: depend on INFINIBAND_ADDR_TRANS

CIFS_SMB_DIRECT code depends on INFINIBAND_ADDR_TRANS provided symbols.
So declare the kconfig dependency.  This is necessary to allow for
enabling INFINIBAND without INFINIBAND_ADDR_TRANS.

Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Reviewed-by: Long Li <longli@microsoft.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoib_srpt: depend on INFINIBAND_ADDR_TRANS
Greg Thelen [Thu, 26 Apr 2018 18:19:32 +0000 (11:19 -0700)]
ib_srpt: depend on INFINIBAND_ADDR_TRANS

INFINIBAND_SRPT code depends on INFINIBAND_ADDR_TRANS provided symbols.
So declare the kconfig dependency.  This is necessary to allow for
enabling INFINIBAND without INFINIBAND_ADDR_TRANS.

Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agonvmet-rdma: depend on INFINIBAND_ADDR_TRANS
Greg Thelen [Thu, 26 Apr 2018 18:19:31 +0000 (11:19 -0700)]
nvmet-rdma: depend on INFINIBAND_ADDR_TRANS

NVME_TARGET_RDMA code depends on INFINIBAND_ADDR_TRANS provided symbols.
So declare the kconfig dependency.  This is necessary to allow for
enabling INFINIBAND without INFINIBAND_ADDR_TRANS.

Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agonvme: depend on INFINIBAND_ADDR_TRANS
Greg Thelen [Thu, 26 Apr 2018 18:19:30 +0000 (11:19 -0700)]
nvme: depend on INFINIBAND_ADDR_TRANS

NVME_RDMA code depends on INFINIBAND_ADDR_TRANS provided symbols.  So
declare the kconfig dependency.  This is necessary to allow for enabling
INFINIBAND without INFINIBAND_ADDR_TRANS.

Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/mlx5: Properly check return value of mlx5_get_uars_page
Leon Romanovsky [Mon, 23 Apr 2018 14:01:56 +0000 (17:01 +0300)]
RDMA/mlx5: Properly check return value of mlx5_get_uars_page

Starting from commit 72f36be06138 ("net/mlx5: Fix mlx5_get_uars_page to
return error code") the mlx5_get_uars_page() call returns error in case
of failure, but it was mistakenly overlooked in the merge commit.

Fixes: e7996a9a77fc ("Merge tag v4.15 of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git")
Reported-by: Alaa Hleihel <alaa@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/mlx5: Fix represent correct netdevice in dual port RoCE
Parav Pandit [Mon, 23 Apr 2018 14:01:55 +0000 (17:01 +0300)]
IB/mlx5: Fix represent correct netdevice in dual port RoCE

In commit bcf87f1dbbec ("IB/mlx5: Listen to netdev register/unresiter events in switchdev mode")
incorrectly mapped primary device's netdevice to 2nd port netdevice.
It always represented primary port's netdevice for 2nd port netdevice
when ib representors were not used.

This results into failing to process CM request arriving on 2nd port due
to incorrect mapping of netdevice.

This fix corrects it by considering the right mdev.

Cc: <stable@vger.kernel.org> # 4.16
Fixes: bcf87f1dbbec ("IB/mlx5: Listen to netdev register/unresiter events in switchdev mode")
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/mlx5: Use unlimited rate when static rate is not supported
Danit Goldberg [Mon, 23 Apr 2018 14:01:54 +0000 (17:01 +0300)]
IB/mlx5: Use unlimited rate when static rate is not supported

Before the change, if the user passed a static rate value different
than zero and the FW doesn't support static rate,
it would end up configuring rate of 2.5 GBps.

Fix this by using rate 0; unlimited, in cases where FW
doesn't support static rate configuration.

Cc: <stable@vger.kernel.org> # 3.10
Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Danit Goldberg <danitg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/mlx5: Protect from shift operand overflow
Leon Romanovsky [Mon, 23 Apr 2018 14:01:53 +0000 (17:01 +0300)]
RDMA/mlx5: Protect from shift operand overflow

Ensure that user didn't supply values too large that can cause overflow.

UBSAN: Undefined behaviour in drivers/infiniband/hw/mlx5/qp.c:263:23
shift exponent -2147483648 is negative
CPU: 0 PID: 292 Comm: syzkaller612609 Not tainted 4.16.0-rc1+ #131
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.0-0-g63451fca13-prebuilt.qemu-project.org 04/01/2014 Call
Trace:
dump_stack+0xde/0x164
ubsan_epilogue+0xe/0x81
set_rq_size+0x7c2/0xa90
create_qp_common+0xc18/0x43c0
mlx5_ib_create_qp+0x379/0x1ca0
create_qp.isra.5+0xc94/0x2260
ib_uverbs_create_qp+0x21b/0x2a0
ib_uverbs_write+0xc2c/0x1010
vfs_write+0x1b0/0x550
SyS_write+0xc7/0x1a0
do_syscall_64+0x1aa/0x740
entry_SYSCALL_64_after_hwframe+0x26/0x9b
RIP: 0033:0x433569
RSP: 002b:00007ffc6e62f448 EFLAGS: 00000217 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 00000000004002f8 RCX: 0000000000433569
RDX: 0000000000000070 RSI: 00000000200042c0 RDI: 0000000000000003
RBP: 00000000006d5018 R08: 00000000004002f8 R09: 00000000004002f8
R10: 00000000004002f8 R11: 0000000000000217 R12: 0000000000000000
R13: 000000000040c9f0 R14: 000000000040ca80 R15: 0000000000000006

Cc: <stable@vger.kernel.org> # 3.10
Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Cc: syzkaller <syzkaller@googlegroups.com>
Reported-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/mlx5: Fix multiple NULL-ptr deref errors in rereg_mr flow
Leon Romanovsky [Mon, 23 Apr 2018 14:01:52 +0000 (17:01 +0300)]
RDMA/mlx5: Fix multiple NULL-ptr deref errors in rereg_mr flow

Failure in rereg MR releases UMEM but leaves the MR to be destroyed
by the user. As a result the following scenario may happen:
"create MR -> rereg MR with failure -> call to rereg MR again" and
hit "NULL-ptr deref or user memory access" errors.

Ensure that rereg MR is only performed on a non-dead MR.

Cc: syzkaller <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org> # 4.5
Fixes: 395a8e4c32ea ("IB/mlx5: Refactoring register MR code")
Reported-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Fix deleting default GIDs when changing mac adddress
Parav Pandit [Mon, 23 Apr 2018 13:58:19 +0000 (16:58 +0300)]
IB/core: Fix deleting default GIDs when changing mac adddress

Before [1], When MAC address of the netdevice is changed, default GID is
supposed to get deleted and added back which affects the node and/or port
GUID in below sequence.

netdevice_event()
-> NETDEV_CHANGEADDR
   default_del_cmd()
      del_netdev_default_ips()
          bond_delete_netdev_default_gids()
              ib_cache_gid_set_default_gid()
                  ib_cache_gid_del()
   add_cmd()
   [..]

However, ib_cache_gid_del() was not getting invoked in non bonding
scenarios because event_ndev and rdma_ndev are same.
Therefore, fix such condition to ignore checking upper device when event
ndev and rdma_dev are same; similar to bond_set_netdev_default_gids().

Which this fix ib_cache_gid_del() is invoked correctly; however
ib_cache_gid_del() doesn't find the default GID for deletion because
find_gid() was given default_gid = false with
GID_ATTR_FIND_MASK_DEFAULT set.
But it was getting overwritten by ib_cache_gid_set_default_gid() later
on as part of add_cmd().
Therefore, mac address change used to work for default GID.

With refactor series [1], this incorrect behavior is detected.

Therefore,
when deleting default GID, set default_gid and set MASK flag.
when deleting IP based GID, clear default_gid and set MASK flag.

[1] https://patchwork.kernel.org/patch/10319151/

Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support")
Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Fix to avoid deleting IPv6 look alike default GIDs
Parav Pandit [Mon, 23 Apr 2018 13:58:18 +0000 (16:58 +0300)]
IB/core: Fix to avoid deleting IPv6 look alike default GIDs

When IPv6 link local address is removed, if it matches with the default
GID, default GID(s)s gets removed which may not be a desired behavior.
This behavior is introduced by refactor work in Fixes tag.

When IPv6 link address is removed, removing its equivalent RoCEv2 GID
which exactly matches with default RoCEv2 GID, is right thing to do.
However achieving it correctly requires lot more changes, likely in
roce_gid_mgmt.c and core/cache.c. This should be done as independent
patch.

Therefore, this patch preserves behavior of not deleteing default GIDs.
This is done by providing explicit hint to consider default GID property
using mask and default_gid; similar to add_gid().

Fixes: 598ff6bae68 ("IB/core: Refactor GID modify code for RoCE")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/core: Don't allow default GID addition at non reseved slots
Parav Pandit [Mon, 23 Apr 2018 13:58:17 +0000 (16:58 +0300)]
IB/core: Don't allow default GID addition at non reseved slots

Default GIDs are marked reserved at the start of the GID table at index
0 and 1 by gid_table_reserve_default().  Currently when default GID is
requested, it can still allocates an empty slot which was not marked as
RESERVED for default GID, which is incorrect.

At least in current code flow of roce_gid_mgmt.c, in theory we can
still request to allocate more than one/two default GIDs depending
on how upper devices are setup.

Therefore, it is better for cache layer to only allow our reserved slots
to be used by default GID allocation requests.

Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agouapi: Fix SPDX tags for files referring to the 'OpenIB.org' license
Jason Gunthorpe [Fri, 20 Apr 2018 15:49:10 +0000 (09:49 -0600)]
uapi: Fix SPDX tags for files referring to the 'OpenIB.org' license

Based on discussion with Kate Stewart this license is not a
BSD-2-Clause, but is now formally identified as Linux-OpenIB
by SPDX.

The key difference between the licenses is in the 'warranty'
paragraph.

if_infiniband.h refers to the 'OpenIB.org' license, but
does not include the text, instead it links to an obsolete
web site that contains a license that matches the BSD-2-Clause
SPX. There is no 'three clause' version of the OpenIB.org
license.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/ucma: Allow resolving address w/o specifying source address
Roland Dreier [Thu, 19 Apr 2018 15:28:11 +0000 (08:28 -0700)]
RDMA/ucma: Allow resolving address w/o specifying source address

The RDMA CM will select a source device and address by consulting
the routing table if no source address is passed into
rdma_resolve_address().  Userspace will ask for this by passing an
all-zero source address in the RESOLVE_IP command.  Unfortunately
the new check for non-zero address size rejects this with EINVAL,
which breaks valid userspace applications.

Fix this by explicitly allowing a zero address family for the source.

Fixes: 2975d5de6428 ("RDMA/ucma: Check AF family prior resolving address")
Cc: <stable@vger.kernel.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoRDMA/ucma: Check for a cm_id->device in all user calls that need it
Jason Gunthorpe [Thu, 5 Apr 2018 03:00:01 +0000 (21:00 -0600)]
RDMA/ucma: Check for a cm_id->device in all user calls that need it

This is done by auditing all callers of ucma_get_ctx and switching the
ones that unconditionally touch ->device to ucma_get_ctx_dev. This covers
a little less than  half of the call sites.

The 11 remaining call sites to ucma_get_ctx() were manually audited.

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: replace refcount_inc with skb_get
Zhu Yanjun [Tue, 10 Apr 2018 13:37:39 +0000 (09:37 -0400)]
IB/rxe: replace refcount_inc with skb_get

Follow the advice from Bart, the function refcount_inc is replaced
with skb_get in commit 99dae690255e ("IB/rxe: optimize mcast recv process")
and commit 86af61764151 ("IB/rxe: remove unnecessary skb_clone").

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Suggested-by: Bart Van Assche <Bart.VanAssche@wdc.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: optimize the function duplicate_request
Zhu Yanjun [Tue, 10 Apr 2018 04:47:15 +0000 (00:47 -0400)]
IB/rxe: optimize the function duplicate_request

In the function duplicate_request, the reference of skb can be increased
to replace the function skb_clone.

This will make rxe performace better and save memory.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/rxe: make rxe_release_udp_tunnel static
Zhu Yanjun [Mon, 9 Apr 2018 12:19:18 +0000 (08:19 -0400)]
IB/rxe: make rxe_release_udp_tunnel static

The function rxe_release_udp_tunnel is only used in rxe_net.c.
So it is necessary to make this function as static.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
6 years agoIB/uverbs: Add missing braces in anonymous union initializers
Geert Uytterhoeven [Mon, 9 Apr 2018 14:52:47 +0000 (16:52 +0200)]
IB/uverbs: Add missing braces in anonymous union initializers

With gcc-4.1.2:

    drivers/infiniband/core/uverbs_std_types_flow_action.c:366: error: unknown field â€˜ptr’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:367: error: unknown field â€˜type’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:367: warning: missing braces around initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:367: warning: (near initialization for â€˜uverbs_flow_action_esp_keymat[0].<anonymous>.<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:368: error: unknown field â€˜min_len’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:368: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:368: warning: (near initialization for â€˜uverbs_flow_action_esp_keymat[0].<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:368: error: unknown field â€˜len’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:368: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:368: warning: (near initialization for â€˜uverbs_flow_action_esp_keymat[0].<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:369: error: unknown field â€˜flags’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:369: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:369: warning: (near initialization for â€˜uverbs_flow_action_esp_keymat[0].<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:376: error: unknown field â€˜ptr’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:377: error: unknown field â€˜type’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:377: warning: missing braces around initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:377: warning: (near initialization for â€˜uverbs_flow_action_esp_replay[0].<anonymous>.<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:379: error: unknown field â€˜len’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:379: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:379: warning: (near initialization for â€˜uverbs_flow_action_esp_replay[0].<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:383: error: unknown field â€˜ptr’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:384: error: unknown field â€˜type’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:385: error: unknown field â€˜min_len’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:385: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:385: warning: (near initialization for â€˜uverbs_flow_action_esp_replay[1].<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:385: error: unknown field â€˜len’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:385: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:385: warning: (near initialization for â€˜uverbs_flow_action_esp_replay[1].<anonymous>’)
    drivers/infiniband/core/uverbs_std_types_flow_action.c:386: error: unknown field â€˜flags’ specified in initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:386: warning: excess elements in union initializer
    drivers/infiniband/core/uverbs_std_types_flow_action.c:386: warning: (near initialization for â€˜uverbs_flow_action_esp_replay[1].<anonymous>’)

Add the missing braces to fix this.

Fixes: 2eb9beaee5d7 ("IB/uverbs: Add flow_action create and destroy verbs")
Fixes: 7d12f8d5a164 ("IB/uverbs: Add modify ESP flow_action")
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoinfiniband: i40iw: Replace GFP_ATOMIC with GFP_KERNEL in i40iw_l2param_change
Jia-Ju Bai [Wed, 11 Apr 2018 07:33:06 +0000 (15:33 +0800)]
infiniband: i40iw: Replace GFP_ATOMIC with GFP_KERNEL in i40iw_l2param_change

i40iw_l2param_change() is never called in atomic context.

i40iw_make_listen_node() is only set as ".l2_param_change"
in struct i40e_client_ops, and this function pointer is not called
in atomic context.

Despite never getting called from atomic context,
i40iw_l2param_change() calls kzalloc() with GFP_ATOMIC,
which does not sleep for allocation.
GFP_ATOMIC is not necessary and can be replaced with GFP_KERNEL,
which can sleep and improve the possibility of sucessful allocation.

This is found by a static analysis tool named DCNS written by myself.
And I also manually check it.

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoinfiniband: i40iw: Replace GFP_ATOMIC with GFP_KERNEL in i40iw_make_listen_node
Jia-Ju Bai [Wed, 11 Apr 2018 07:32:48 +0000 (15:32 +0800)]
infiniband: i40iw: Replace GFP_ATOMIC with GFP_KERNEL in i40iw_make_listen_node

i40iw_make_listen_node() is never called in atomic context.

i40iw_make_listen_node() is only called by i40iw_create_listen, which is
set as ".create_listen" in struct iw_cm_verbs.

Despite never getting called from atomic context,
i40iw_make_listen_node() calls kzalloc() with GFP_ATOMIC,
which does not sleep for allocation.
GFP_ATOMIC is not necessary and can be replaced with GFP_KERNEL,
which can sleep and improve the possibility of sucessful allocation.

This is found by a static analysis tool named DCNS written by myself.
And I also manually check it.

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoinfiniband: i40iw: Replace GFP_ATOMIC with GFP_KERNEL in i40iw_add_mqh_4
Jia-Ju Bai [Wed, 11 Apr 2018 07:32:25 +0000 (15:32 +0800)]
infiniband: i40iw: Replace GFP_ATOMIC with GFP_KERNEL in i40iw_add_mqh_4

i40iw_add_mqh_4() is never called in atomic context, because it
calls rtnl_lock() that can sleep.

Despite never getting called from atomic context,
i40iw_add_mqh_4() calls kzalloc() with GFP_ATOMIC,
which does not sleep for allocation.
GFP_ATOMIC is not necessary and can be replaced with GFP_KERNEL,
which can sleep and improve the possibility of sucessful allocation.

This is found by a static analysis tool named DCNS written by myself.
And I also manually check it.

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoIB/rxe: avoid export symbols
Zhu Yanjun [Sat, 14 Apr 2018 11:44:44 +0000 (07:44 -0400)]
IB/rxe: avoid export symbols

The functions rxe_set_mtu, rxe_add and rxe_remove are only used in their
own module. So it is not necessary to export them.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoIB/rxe: make the variable static
Zhu Yanjun [Wed, 4 Apr 2018 00:08:20 +0000 (20:08 -0400)]
IB/rxe: make the variable static

The variable rxe_net_notifier is only used in the file rxe_net.c. So
remove it from rxe_net.h file and make it static in the file rxe_net.c.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoRDMA/rdma_cm: Delete rdma_addr_client
Jason Gunthorpe [Tue, 3 Apr 2018 04:52:04 +0000 (07:52 +0300)]
RDMA/rdma_cm: Delete rdma_addr_client

The only thing it does is block module unload while work is posted from
rdma_resolve_ip().

However, this is not the right place to do this. The users of
rdma_resolve_ip() must ensure their own module does not unload until
rdma_resolve_ip() calls the callback, or until rdma_addr_cancel() is
called.

Similarly callers to rdma_addr_find_l2_eth_by_grh() must ensure their
module does not unload while they are calling code.

The only two users are already safe, so there is no need for this.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoRDMA/rdma_cm: Make rdma_addr_cancel into a fence
Jason Gunthorpe [Tue, 3 Apr 2018 04:52:03 +0000 (07:52 +0300)]
RDMA/rdma_cm: Make rdma_addr_cancel into a fence

Currently rdma_addr_cancel does not prevent the callback from being used,
this is surprising and hard to reason about. There does not appear to be a
bug here as the only user of this API does refcount properly, fixing it
only to increase clarity.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoRDMA/rdma_cm: Remove process_req and timer sorting
Jason Gunthorpe [Tue, 3 Apr 2018 04:52:02 +0000 (07:52 +0300)]
RDMA/rdma_cm: Remove process_req and timer sorting

Now that the work queue is used directly to launch and track the work
there is no need for the second processing function to do 'all list
entries'. Just schedule all entries onto the main work queue directly.

We can also drop all of the useless list sorting now, as the workqueue
sorts by expiration time automatically.

This change requires switching lock to a spinlock as netdev notifiers
are called in an atomic context, this is now easy since the lock does
not need to be held across the lookup, that is already single
threaded due to the work queue.

Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoinfiniband: mlx5: fix build errors when INFINIBAND_USER_ACCESS=m
Randy Dunlap [Tue, 17 Apr 2018 01:51:50 +0000 (18:51 -0700)]
infiniband: mlx5: fix build errors when INFINIBAND_USER_ACCESS=m

Fix build errors when INFINIBAND_USER_ACCESS=m and MLX5_INFINIBAND=y.
The build error occurs when the mlx5 driver code attempts to use
USER_ACCESS interfaces, which are built as a loadable module.

Fixes these build errors:

drivers/infiniband/hw/mlx5/main.o: In function `populate_specs_root':
../drivers/infiniband/hw/mlx5/main.c:4982: undefined reference to `uverbs_default_get_objects'
../drivers/infiniband/hw/mlx5/main.c:4994: undefined reference to `uverbs_alloc_spec_tree'
drivers/infiniband/hw/mlx5/main.o: In function `depopulate_specs_root':
../drivers/infiniband/hw/mlx5/main.c:5001: undefined reference to `uverbs_free_spec_tree'

Build-tested with multiple config combinations.

Fixes: 8c84660bb437 ("IB/mlx5: Initialize the parsing tree root without the help of uverbs")
Cc: stable@vger.kernel.org # reported against 4.16
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoIB/mlx5: remove duplicate header file
Zhu Yanjun [Sun, 15 Apr 2018 01:16:54 +0000 (21:16 -0400)]
IB/mlx5: remove duplicate header file

The header file fs_helpers.h is included twice. So it should be removed.

Fixes: 802c2125689d ("IB/mlx5: Add IPsec support for egress and ingress")
CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoRDMA/ucma: ucma_context reference leak in error path
Shamir Rabinovitch [Tue, 10 Apr 2018 14:26:23 +0000 (10:26 -0400)]
RDMA/ucma: ucma_context reference leak in error path

Validating input parameters should be done before getting the cm_id
otherwise it can leak a cm_id reference.

Fixes: 6a21dfc0d0db ("RDMA/ucma: Limit possible option size")
Signed-off-by: Shamir Rabinovitch <shamir.rabinovitch@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
6 years agoLinux 4.17-rc1
Linus Torvalds [Mon, 16 Apr 2018 01:24:20 +0000 (18:24 -0700)]
Linux 4.17-rc1

6 years agoMerge tag 'for-4.17-part2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
Linus Torvalds [Mon, 16 Apr 2018 01:08:35 +0000 (18:08 -0700)]
Merge tag 'for-4.17-part2-tag' of git://git./linux/kernel/git/kdave/linux

Pull more btrfs updates from David Sterba:
 "We have queued a few more fixes (error handling, log replay,
  softlockup) and the rest is SPDX updates that touche almost all files
  so the diffstat is long"

* tag 'for-4.17-part2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: Only check first key for committed tree blocks
  btrfs: add SPDX header to Kconfig
  btrfs: replace GPL boilerplate by SPDX -- sources
  btrfs: replace GPL boilerplate by SPDX -- headers
  Btrfs: fix loss of prealloc extents past i_size after fsync log replay
  Btrfs: clean up resources during umount after trans is aborted
  btrfs: Fix possible softlock on single core machines
  Btrfs: bail out on error during replay_dir_deletes
  Btrfs: fix NULL pointer dereference in log_dir_items

6 years agoMerge tag '4.17-rc1SMB3-Fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Mon, 16 Apr 2018 01:06:22 +0000 (18:06 -0700)]
Merge tag '4.17-rc1SMB3-Fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "SMB3 fixes, a few for stable, and some important cleanup work from
  Ronnie of the smb3 transport code"

* tag '4.17-rc1SMB3-Fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: change validate_buf to validate_iov
  cifs: remove rfc1002 hardcoded constants from cifs_discard_remaining_data()
  cifs: Change SMB2_open to return an iov for the error parameter
  cifs: add resp_buf_size to the mid_q_entry structure
  smb3.11: replace a 4 with server->vals->header_preamble_size
  cifs: replace a 4 with server->vals->header_preamble_size
  cifs: add pdu_size to the TCP_Server_Info structure
  SMB311: Improve checking of negotiate security contexts
  SMB3: Fix length checking of SMB3.11 negotiate request
  CIFS: add ONCE flag for cifs_dbg type
  cifs: Use ULL suffix for 64-bit constant
  SMB3: Log at least once if tree connect fails during reconnect
  cifs: smb2pdu: Fix potential NULL pointer dereference

6 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Mon, 16 Apr 2018 00:24:12 +0000 (17:24 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "This is a set of minor (and safe changes) that didn't make the initial
  pull request plus some bug fixes.

  The status handling code is actually a running regression from the
  previous merge window which had an incomplete fix (now reverted) and
  most of the remaining bug fixes are for problems older than the
  current merge window"

[ Side note: this merge also takes the base kernel git repository to 6+
  million objects for the first time. Technically we hit it a couple of
  merges ago already if you count all the tag objects, but now it
  reaches 6M+ objects reachable from HEAD.

  I was joking around that that's when I should switch to 5.0, because
  3.0 happened at the 2M mark, and 4.0 happened at 4M objects. But
  probably not, even if numerology is about as good a reason as any.

                                                              - Linus ]

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: devinfo: Add Microsoft iSCSI target to 1024 sector blacklist
  scsi: cxgb4i: silence overflow warning in t4_uld_rx_handler()
  scsi: dpt_i2o: Use after free in I2ORESETCMD ioctl
  scsi: core: Make scsi_result_to_blk_status() recognize CONDITION MET
  scsi: core: Rename __scsi_error_from_host_byte() into scsi_result_to_blk_status()
  Revert "scsi: core: return BLK_STS_OK for DID_OK in __scsi_error_from_host_byte()"
  scsi: aacraid: Insure command thread is not recursively stopped
  scsi: qla2xxx: Correct setting of SAM_STAT_CHECK_CONDITION
  scsi: qla2xxx: correctly shift host byte
  scsi: qla2xxx: Fix race condition between iocb timeout and initialisation
  scsi: qla2xxx: Avoid double completion of abort command
  scsi: qla2xxx: Fix small memory leak in qla2x00_probe_one on probe failure
  scsi: scsi_dh: Don't look for NULL devices handlers by name
  scsi: core: remove redundant assignment to shost->use_blk_mq

6 years agoMerge tag 'kbuild-v4.17-2' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy...
Linus Torvalds [Mon, 16 Apr 2018 00:21:30 +0000 (17:21 -0700)]
Merge tag 'kbuild-v4.17-2' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull more Kbuild updates from Masahiro Yamada:

 - pass HOSTLDFLAGS when compiling single .c host programs

 - build genksyms lexer and parser files instead of using shipped
   versions

 - rename *-asn1.[ch] to *.asn1.[ch] for suffix consistency

 - let the top .gitignore globally ignore artifacts generated by flex,
   bison, and asn1_compiler

 - let the top Makefile globally clean artifacts generated by flex,
   bison, and asn1_compiler

 - use safer .SECONDARY marker instead of .PRECIOUS to prevent
   intermediate files from being removed

 - support -fmacro-prefix-map option to make __FILE__ a relative path

 - fix # escaping to prepare for the future GNU Make release

 - clean up deb-pkg by using debian tools instead of handrolled
   source/changes generation

 - improve rpm-pkg portability by supporting kernel-install as a
   fallback of new-kernel-pkg

 - extend Kconfig listnewconfig target to provide more information

* tag 'kbuild-v4.17-2' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kconfig: extend output of 'listnewconfig'
  kbuild: rpm-pkg: use kernel-install as a fallback for new-kernel-pkg
  Kbuild: fix # escaping in .cmd files for future Make
  kbuild: deb-pkg: split generating packaging and build
  kbuild: use -fmacro-prefix-map to make __FILE__ a relative path
  kbuild: mark $(targets) as .SECONDARY and remove .PRECIOUS markers
  kbuild: rename *-asn1.[ch] to *.asn1.[ch]
  kbuild: clean up *-asn1.[ch] patterns from top-level Makefile
  .gitignore: move *-asn1.[ch] patterns to the top-level .gitignore
  kbuild: add %.dtb.S and %.dtb to 'targets' automatically
  kbuild: add %.lex.c and %.tab.[ch] to 'targets' automatically
  genksyms: generate lexer and parser during build instead of shipping
  kbuild: clean up *.lex.c and *.tab.[ch] patterns from top-level Makefile
  .gitignore: move *.lex.c *.tab.[ch] patterns to the top-level .gitignore
  kbuild: use HOSTLDFLAGS for single .c executables

6 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Apr 2018 23:12:35 +0000 (16:12 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A set of fixes and updates for x86:

   - Address a swiotlb regression which was caused by the recent DMA
     rework and made driver fail because dma_direct_supported() returned
     false

   - Fix a signedness bug in the APIC ID validation which caused invalid
     APIC IDs to be detected as valid thereby bloating the CPU possible
     space.

   - Fix inconsisten config dependcy/select magic for the MFD_CS5535
     driver.

   - Fix a corruption of the physical address space bits when encryption
     has reduced the address space and late cpuinfo updates overwrite
     the reduced bit information with the original value.

   - Dominiks syscall rework which consolidates the architecture
     specific syscall functions so all syscalls can be wrapped with the
     same macros. This allows to switch x86/64 to struct pt_regs based
     syscalls. Extend the clearing of user space controlled registers in
     the entry patch to the lower registers"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/apic: Fix signedness bug in APIC ID validity checks
  x86/cpu: Prevent cpuinfo_x86::x86_phys_bits adjustment corruption
  x86/olpc: Fix inconsistent MFD_CS5535 configuration
  swiotlb: Use dma_direct_supported() for swiotlb_ops
  syscalls/x86: Adapt syscall_wrapper.h to the new syscall stub naming convention
  syscalls/core, syscalls/x86: Rename struct pt_regs-based sys_*() to __x64_sys_*()
  syscalls/core, syscalls/x86: Clean up compat syscall stub naming convention
  syscalls/core, syscalls/x86: Clean up syscall stub naming convention
  syscalls/x86: Extend register clearing on syscall entry to lower registers
  syscalls/x86: Unconditionally enable 'struct pt_regs' based syscalls on x86_64
  syscalls/x86: Use 'struct pt_regs' based syscall calling for IA32_EMULATION and x32
  syscalls/core: Prepare CONFIG_ARCH_HAS_SYSCALL_WRAPPER=y for compat syscalls
  syscalls/x86: Use 'struct pt_regs' based syscall calling convention for 64-bit syscalls
  syscalls/core: Introduce CONFIG_ARCH_HAS_SYSCALL_WRAPPER=y
  x86/syscalls: Don't pointlessly reload the system call number
  x86/mm: Fix documentation of module mapping range with 4-level paging
  x86/cpuid: Switch to 'static const' specifier

6 years agoMerge branch 'x86-pti-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 15 Apr 2018 20:35:29 +0000 (13:35 -0700)]
Merge branch 'x86-pti-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 pti updates from Thomas Gleixner:
 "Another series of PTI related changes:

   - Remove the manual stack switch for user entries from the idtentry
     code. This debloats entry by 5k+ bytes of text.

   - Use the proper types for the asm/bootparam.h defines to prevent
     user space compile errors.

   - Use PAGE_GLOBAL for !PCID systems to gain back performance

   - Prevent setting of huge PUD/PMD entries when the entries are not
     leaf entries otherwise the entries to which the PUD/PMD points to
     and are populated get lost"

* 'x86-pti-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/pgtable: Don't set huge PUD/PMD on non-leaf entries
  x86/pti: Leave kernel text global for !PCID
  x86/pti: Never implicitly clear _PAGE_GLOBAL for kernel image
  x86/pti: Enable global pages for shared areas
  x86/mm: Do not forbid _PAGE_RW before init for __ro_after_init
  x86/mm: Comment _PAGE_GLOBAL mystery
  x86/mm: Remove extra filtering in pageattr code
  x86/mm: Do not auto-massage page protections
  x86/espfix: Document use of _PAGE_GLOBAL
  x86/mm: Introduce "default" kernel PTE mask
  x86/mm: Undo double _PAGE_PSE clearing
  x86/mm: Factor out pageattr _PAGE_GLOBAL setting
  x86/entry/64: Drop idtentry's manual stack switch for user entries
  x86/uapi: Fix asm/bootparam.h userspace compilation errors

6 years agoMerge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Apr 2018 19:43:30 +0000 (12:43 -0700)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Thomas Gleixner:
 "A few scheduler fixes:

   - Prevent a bogus warning vs. runqueue clock update flags in
     do_sched_rt_period_timer()

   - Simplify the helper functions which handle requests for skipping
     the runqueue clock updat.

   - Do not unlock the tunables mutex in the error path of the cpu
     frequency scheduler utils. Its not held.

   - Enforce proper alignement for 'struct util_est' in sched_avg to
     prevent a misalignment fault on IA64"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/core: Force proper alignment of 'struct util_est'
  sched/core: Simplify helpers for rq clock update skip requests
  sched/rt: Fix rq->clock_update_flags < RQCF_ACT_SKIP warning
  sched/cpufreq/schedutil: Fix error path mutex unlock

6 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Apr 2018 19:36:31 +0000 (12:36 -0700)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull more perf updates from Thomas Gleixner:
 "A rather large set of perf updates:

  Kernel:

   - Fix various initialization issues

   - Prevent creating [ku]probes for not CAP_SYS_ADMIN users

  Tooling:

   - Show only failing syscalls with 'perf trace --failure' (Arnaldo
     Carvalho de Melo)

            e.g: See what 'openat' syscalls are failing:

        # perf trace --failure -e openat
         762.323 ( 0.007 ms): VideoCapture/4566 openat(dfd: CWD, filename: /dev/video2) = -1 ENOENT No such file or directory
         <SNIP N /dev/videoN open attempts... sigh, where is that improvised camera lid?!? >
         790.228 ( 0.008 ms): VideoCapture/4566 openat(dfd: CWD, filename: /dev/video63) = -1 ENOENT No such file or directory
        ^C#

   - Show information about the event (freq, nr_samples, total
     period/nr_events) in the annotate --tui and --stdio2 'perf
     annotate' output, similar to the first line in the 'perf report
     --tui', but just for the samples for a the annotated symbol
     (Arnaldo Carvalho de Melo)

   - Introduce 'perf version --build-options' to show what features were
     linked, aliased as well as a shorter 'perf -vv' (Jin Yao)

   - Add a "dso_size" sort order (Kim Phillips)

   - Remove redundant ')' in the tracepoint output in 'perf trace'
     (Changbin Du)

   - Synchronize x86's cpufeatures.h, no effect on toolss (Arnaldo
     Carvalho de Melo)

   - Show group details on the title line in the annotate browser and
     'perf annotate --stdio2' output, so that the per-event columns can
     have headers (Arnaldo Carvalho de Melo)

   - Fixup vertical line separating metrics from instructions and
     cleaning unused lines at the bottom, both in the annotate TUI
     browser (Arnaldo Carvalho de Melo)

   - Remove duplicated 'samples' in lost samples warning in
     'perf report' (Arnaldo Carvalho de Melo)

   - Synchronize i915_drm.h, silencing the perf build process,
     automagically adding support for the new DRM_I915_QUERY ioctl
     (Arnaldo Carvalho de Melo)

   - Make auxtrace_queues__add_buffer() allocate struct buffer, from a
     patchkit already applied (Adrian Hunter)

   - Fix the --stdio2/TUI annotate output to include group details, be
     it for a recorded '{a,b,f}' explicit event group or when forcing
     group display using 'perf report --group' for a set of events not
     recorded as a group (Arnaldo Carvalho de Melo)

   - Fix display artifacts in the ui browser (base class for the
     annotate and main report/top TUI browser) related to the extra
     title lines work (Arnaldo Carvalho de Melo)

   - perf auxtrace refactorings, leftovers from a previously partially
     processed patchset (Adrian Hunter)

   - Fix the builtin clang build (Sandipan Das, Arnaldo Carvalho de
     Melo)

   - Synchronize i915_drm.h, silencing a perf build warning and in the
     process automagically adding support for a new ioctl command
     (Arnaldo Carvalho de Melo)

   - Fix a strncpy issue in uprobe tracing"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (36 commits)
  perf/core: Need CAP_SYS_ADMIN to create k/uprobe with perf_event_open()
  tracing/uprobe_event: Fix strncpy corner case
  perf/core: Fix perf_uprobe_init()
  perf/core: Fix perf_kprobe_init()
  perf/core: Fix use-after-free in uprobe_perf_close()
  perf tests clang: Fix function name for clang IR test
  perf clang: Add support for recent clang versions
  perf tools: Fix perf builds with clang support
  perf tools: No need to include namespaces.h in util.h
  perf hists browser: Remove leftover from row returned from refresh
  perf hists browser: Show extra_title_lines in the 'D' debug hotkey
  perf auxtrace: Make auxtrace_queues__add_buffer() do CPU filtering
  tools headers uapi: Synchronize i915_drm.h
  perf report: Remove duplicated 'samples' in lost samples warning
  perf ui browser: Fixup cleaning unused lines at the bottom
  perf annotate browser: Fixup vertical line separating metrics from instructions
  perf annotate: Show group details on the title line
  perf auxtrace: Make auxtrace_queues__add_buffer() allocate struct buffer
  perf/x86/intel: Move regs->flags EXACT bit init
  perf trace: Remove redundant ')'
  ...

6 years agoMerge branch 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Apr 2018 19:32:06 +0000 (12:32 -0700)]
Merge branch 'efi-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 EFI bootup fixlet from Thomas Gleixner:
 "A single fix for an early boot warning caused by invoking
  this_cpu_has() before SMP initialization"

* 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mm: Fix bogus warning during EFI bootup, use boot_cpu_has() instead of this_cpu_has() in build_cr3_noflush()

6 years agoMerge branch 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 15 Apr 2018 19:29:46 +0000 (12:29 -0700)]
Merge branch 'irq-core-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq affinity fixes from Thomas Gleixner:

  - Fix error path handling in the affinity spreading code

  - Make affinity spreading smarter to avoid issues on systems which
    claim to have hotpluggable CPUs while in fact they can't hotplug
    anything.

    So instead of trying to spread the vectors (and thereby the
    associated device queues) to all possibe CPUs, spread them on all
    present CPUs first. If there are left over vectors after that first
    step they are spread among the possible, but not present CPUs which
    keeps the code backwards compatible for virtual decives and NVME
    which allocate a queue per possible CPU, but makes the spreading
    smarter for devices which have less queues than possible or present
    CPUs.

* 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  genirq/affinity: Spread irq vectors among present CPUs as far as possible
  genirq/affinity: Allow irq spreading from a given starting point
  genirq/affinity: Move actual irq vector spreading into a helper function
  genirq/affinity: Rename *node_to_possible_cpumask as *node_to_cpumask
  genirq/affinity: Don't return with empty affinity masks on error