linux-stable.git/drivers/infiniband/ulp, branch v4.7.2

IB/srpt: Limit the number of SG elements per work request

2016-08-20T16:11:02+00:00

commit 30c6d8773de06878f920666d8c945f81cb2081b3 upstream.

Limit the number of SG elements per work request to what the HCA
and the queue pair support.

Fixes: 34693573fde0 ("IB/srpt: Reduce QP buffer size")
Reported-by: Parav Pandit 
Signed-off-by: Bart Van Assche 
Cc: Christoph Hellwig 
Cc: Sagi Grimberg 
Cc: Steve Wise 
Cc: Parav Pandit 
Cc: Nicholas Bellinger 
Cc: Laurence Oberman 
Reviewed-by: Christoph Hellwig 
Signed-off-by: Doug Ledford 
Signed-off-by: Greg Kroah-Hartman

IB/srpt: Reduce QP buffer size

2016-06-23T16:04:09+00:00

The memory needed for the send and receive queues associated with
a QP is proportional to the max_sge parameter. The current value
of that parameter is such that with an mlx4 HCA the QP buffer size
is 8 MB. Since DMA is used for communication between HCA and CPU
that buffer either has to be allocated coherently or map_single()
must succeed for that buffer. Since large contiguous allocations
are fragile and since the maximum segment size for e.g. swiotlb
is 256 KB, reduce the max_sge parameter. This patch avoids that
the following text appears on the console after SRP logout and
relogin on a system equipped with multiple IB HCAs:

mlx4_core 0000:05:00.0: swiotlb buffer is full (sz: 8388608 bytes)
swiotlb: coherent allocation failed for device 0000:05:00.0 size=8388608
CPU: 11 PID: 148 Comm: kworker/11:1 Not tainted 4.7.0-rc4-dbg+ #1
Call Trace:
 [] dump_stack+0x67/0x92
 [] swiotlb_alloc_coherent+0x141/0x150
 [] x86_swiotlb_alloc_coherent+0x3e/0x50
 [] mlx4_buf_direct_alloc.isra.5+0x9a/0x120 [mlx4_core]
 [] mlx4_buf_alloc+0x165/0x1a0 [mlx4_core]
 [] create_qp_common.isra.29+0x57d/0xff0 [mlx4_ib]
 [] mlx4_ib_create_qp+0x12a/0x3f0 [mlx4_ib]
 [] ib_create_qp+0x3a/0x250 [ib_core]
 [] srpt_cm_handler+0x4bb/0xcad [ib_srpt]
 [] cm_process_work+0x20/0xf0 [ib_cm]
 [] cm_work_handler+0x1ac0/0x2059 [ib_cm]
 [] process_one_work+0x19d/0x490
 [] worker_thread+0x49/0x490
 [] kthread+0xea/0x100
 [] ret_from_fork+0x1f/0x40

Fixes: b99f8e4d7bcd ("IB/srpt: convert to the generic RDMA READ/WRITE API")
Signed-off-by: Bart Van Assche 
Cc: Laurence Oberman 
Cc: Christoph Hellwig 
Cc: Sagi Grimberg 
Signed-off-by: Doug Ledford

IB/IPoIB: Don't update neigh validity for unresolved entries

2016-06-07T14:49:48+00:00

ipoib_neigh_get unconditionally updates the "alive" variable member on
any packet send.  This prevents the neighbor garbage collection from
cleaning out a dead neighbor entry if we are still queueing packets
for it.  If the queue for this neighbor is full, then don't update the
alive timestamp.  That way the neighbor can time out even if packets
are still being queued as long as none of them are being sent.

Fixes: b63b70d87741 ("IPoIB: Use a private hash table for path lookup in xmit path")
Signed-off-by: Erez Shitrit 
Signed-off-by: Leon Romanovsky 
Signed-off-by: Doug Ledford

IB/IPoIB: Disable bottom half when dealing with device address

2016-06-07T13:50:54+00:00

Align locking usage when touching device address with rest
of the kernel. Lock the bottom half when doing so using
netif_addr_lock_bh.

This also solves the following case as reported by lockdep:
	CPU0                    CPU1
	----                    ----
lock(_xmit_INFINIBAND);
				local_irq_disable();
				lock(&(&mc->mca_lock)->rlock);
				lock(_xmit_INFINIBAND);

lock(&(&mc->mca_lock)->rlock);

*** DEADLOCK ***

Fixes: 492a7e67ff83 ("IB/IPoIB: Allow setting the device address")
Signed-off-by: Mark Bloch 
Signed-off-by: Leon Romanovsky 
Signed-off-by: Doug Ledford

IB/IPoIB: Fix race between ipoib_remove_one to sysfs functions

2016-06-07T13:50:53+00:00

In ipoib_remove_one the driver holds the rtnl_lock and tries to do some
operation like dev_change_flags or unregister_netdev, while sysfs
callback like ipoib_vlan_delete holds sysfs mutex and tries to hold the
rtnl_lock via rtnl_trylock() and restart_syscall() if the lock is not
free, meanwhile ipoib_remove_one tries to get the sysfs lock in order to
free its sysfs directory, and we will get  a->b, b->a deadlock.

    Trace like the following:

        schedule+0x37/0x80
        schedule_preempt_disabled+0xe/0x10
        __mutex_lock_slowpath+0xb5/0x120
        mutex_lock+0x23/0x40
        rtnl_lock+0x15/0x20
        netdev_run_todo+0x17c/0x320
        rtnl_unlock+0xe/0x10
        ipoib_vlan_delete+0x11b/0x1b0 [ib_ipoib]
        delete_child+0x54/0x80 [ib_ipoib]
        dev_attr_store+0x18/0x30
        sysfs_kf_write+0x37/0x40
        mutex_lock+0x16/0x40
        SyS_write+0x55/0xc0
        entry_SYSCALL_64_fastpath+0x16/0x75
    And
        schedule+0x37/0x80
        __kernfs_remove+0x1a8/0x260
        ? wake_atomic_t_function+0x60/0x60
        kernfs_remove+0x25/0x40
        sysfs_remove_dir+0x50/0x80
        kobject_del+0x18/0x50
        device_del+0x19f/0x260
        netdev_unregister_kobject+0x6a/0x80
        rollback_registered_many+0x1fd/0x340
        rollback_registered+0x3c/0x70
        unregister_netdevice_queue+0x55/0xc0
        unregister_netdev+0x20/0x30
        ipoib_remove_one+0x114/0x1b0 [ib_ipoib]
        ib_unregister_client+0x4a/0x170 [ib_core]
        ? find_module_all+0x71/0xa0
        ipoib_cleanup_module+0x10/0x94 [ib_ipoib]
        SyS_delete_module+0x1b5/0x210
        entry_SYSCALL_64_fastpath+0x16/0x75

The fix is by checking the flag IPOIB_FLAG_INTF_ON_DESTROY in order to
get out from the sysfs function.

Fixes: 862096a8bbf8 ("IB/ipoib: Add more rtnl_link_ops callbacks")
Fixes: 9baa0b036410 ("IB/ipoib: Add rtnl_link_ops support")
Signed-off-by: Erez Shitrit 
Signed-off-by: Leon Romanovsky 
Signed-off-by: Doug Ledford

IB/srp: Fix srp_map_sg_dma()

2016-06-06T23:32:59+00:00

Because patch "IB/srp: Move common code into the caller" was applied
partially srp_map_sg_dma() doesn't work properly. Fix this by
applying the remainder of that patch. See also
http://thread.gmane.org/gmane.linux.drivers.rdma/35803/focus=35811.

Fixes: 3849e44d1c4b ("IB/srp: Move common code into the caller")
Signed-off-by: Bart Van Assche 
Cc: Mike Marciniszyn 
Cc: Sagi Grimberg 
Cc: Christoph Hellwig 
Cc: Laurence Oberman 
Reviewed-by: Sagi Grimberg 
Reviewed-by: Christoph Hellwig 
Signed-off-by: Doug Ledford

IB/srp: Always initialize use_fast_reg and use_fmr

2016-06-06T23:32:58+00:00

Avoid that mapping fails due to use_fast_reg != 0 or use_fmr != 0
if both member variables should be zero (if never_register == 1 or
if neither FMR nor FR is supported). Remove an initialization that
became superfluous due to changing a kmalloc() into a kzalloc()
call.

Fixes: 509c5f33f4f6 ("IB/srp: Prevent mapping failures")
Cc: Sagi Grimberg 
Cc: Christoph Hellwig 
Cc: Laurence Oberman 
Signed-off-by: Bart Van Assche 
Reviewed-by: Leon Romanovsky 
Reviewed-by: Sagi Grimberg 
Reviewed-by: Christoph Hellwig 
Signed-off-by: Doug Ledford

Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending

2016-05-28T19:04:17+00:00

Pull SCSI target updates from Nicholas Bellinger:
 "Here are the outstanding target pending updates for v4.7-rc1.

  The highlights this round include:

   - Allow external PR/ALUA metadata path be defined at runtime via top
     level configfs attribute (Lee)
   - Fix target session shutdown bug for ib_srpt multi-channel (hch)
   - Make TFO close_session() and shutdown_session() optional (hch)
   - Drop se_sess->sess_kref + convert tcm_qla2xxx to internal kref
     (hch)
   - Add tcm_qla2xxx endpoint attribute for basic FC jammer (Laurence)
   - Refactor iscsi-target RX/TX PDU encode/decode into common code
     (Varun)
   - Extend iscsit_transport with xmit_pdu, release_cmd, get_rx_pdu,
     validate_parameters, and get_r2t_ttt for generic ISO offload
     (Varun)
   - Initial merge of cxgb iscsi-segment offload target driver (Varun)

  The bulk of the changes are Chelsio's new driver, along with a number
  of iscsi-target common code improvements made by Varun + Co along the
  way"

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (29 commits)
  iscsi-target: Fix early sk_data_ready LOGIN_FLAGS_READY race
  cxgbit: Use type ISCSI_CXGBIT + cxgbit tpg_np attribute
  iscsi-target: Convert transport drivers to signal rdma_shutdown
  iscsi-target: Make iscsi_tpg_np driver show/store use generic code
  tcm_qla2xxx Add SCSI command jammer/discard capability
  iscsi-target: graceful disconnect on invalid mapping to iovec
  target: need_to_release is always false, remove redundant check and kfree
  target: remove sess_kref and ->shutdown_session
  iscsi-target: remove usage of ->shutdown_session
  tcm_qla2xxx: introduce a private sess_kref
  target: make close_session optional
  target: make ->shutdown_session optional
  target: remove acl_stop
  target: consolidate and fix session shutdown
  cxgbit: add files for cxgbit.ko
  iscsi-target: export symbols
  iscsi-target: call complete on conn_logout_comp
  iscsi-target: clear tx_thread_active
  iscsi-target: add new offload transport type
  iscsi-target: use conn_transport->transport_type in text rsp
  ...

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

2016-05-28T18:04:16+00:00

Pull more rdma updates from Doug Ledford:
 "This is the second group of code for the 4.7 merge window.  It looks
  large, but only in one sense.  I'll get to that in a minute.  The list
  of changes here breaks down as follows:

   - Dynamic counter infrastructure in the IB drivers

     This is a sysfs based code to allow free form access to the
     hardware counters RDMA devices might support so drivers don't need
     to code this up repeatedly themselves

   - SendOnlyFullMember multicast support

   - IB router support

   - A couple misc fixes

   - The big item on the list: hfi1 driver updates, plus moving the hfi1
     driver out of staging

  There was a group of 15 patches in the hfi1 list that I thought I had
  in the first pull request but they weren't.  So that added to the
  length of the hfi1 section here.

  As far as these go, everything but the hfi1 is pretty straight
  forward.

  The hfi1 is, if you recall, the driver that Al had complaints about
  how it used the write/writev interfaces in an overloaded fashion.  The
  write portion of their interface behaved like the write handler in the
  IB stack proper and did bi-directional communications.  The writev
  interface, on the other hand, only accepts SDMA request structures.
  The completions for those structures are sent back via an entirely
  different event mechanism.

  With the security patch, we put security checks on the write
  interface, however, we also knew they would be going away soon.  Now,
  we've converted the write handler in the hfi1 driver to use ioctls
  from the IB reserved magic area for its bidirectional communications.
  With that change, Intel has addressed all of the items originally on
  their TODO when they went into staging (as well as many items added to
  the list later).

  As such, I moved them out, and since they were the last item in the
  staging/rdma directory, and I don't have immediate plans to use the
  staging area again, I removed the staging/rdma area.

  Because of the move out of staging, as well as a series of 5 patches
  in the hfi1 driver that removed code people thought should be done in
  a different way and was optional to begin with (a snoop debug
  interface, an eeprom driver for an eeprom connected directory to their
  hfi1 chip and not via an i2c bus, and a few other things like that),
  the line count, especially the removal count, is high"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (56 commits)
  staging/rdma: Remove the entire rdma subdirectory of staging
  IB/core: Make device counter infrastructure dynamic
  IB/hfi1: Fix pio map initialization
  IB/hfi1: Correct 8051 link parameter settings
  IB/hfi1: Update pkey table properly after link down or FM start
  IB/rdamvt: Fix rdmavt s_ack_queue sizing
  IB/rdmavt: Max atomic value should be a u8
  IB/hfi1: Fix hard lockup due to not using save/restore spin lock
  IB/hfi1: Add tracing support for send with invalidate opcode
  IB/hfi1, qib: Add ieth to the packet header definitions
  IB/hfi1: Move driver out of staging
  IB/hfi1: Do not free hfi1 cdev parent structure early
  IB/hfi1: Add trace message in user IOCTL handling
  IB/hfi1: Remove write(), use ioctl() for user cmds
  IB/hfi1: Add ioctl() interface for user commands
  IB/hfi1: Remove unused user command
  IB/hfi1: Remove snoop/diag interface
  IB/hfi1: Remove EPROM functionality from data device
  IB/hfi1: Remove UI char device
  IB/hfi1: Remove multiple device cdev
  ...

IB/IPoIB: Allow setting the device address

2016-05-25T19:39:03+00:00

In IB networks, and specifically in IPoIB/rdmacm traffic, the device
address of an IPoIB interface is used as a means to exchange information
between nodes needed for communication.

Currently an IPoIB interface will always be created with a device
address based on its node GUID without a way to change that.

This change adds the ability to set the device address of an IPoIB
interface by value. We use the set mac address ndo to do that.

The flow should be broken down to two:
1) The GID value is already in the GID table,
   in this case the interface will be able to set carrier up.

2) The GID value is not yet in the GID table,
   in this case the interface won't try to join the multicast group
   and will wait (listen on GID_CHANGE event) until the GID is inserted.

In order to track those changes, we add a new flag:
* IPOIB_FLAG_DEV_ADDR_SET.

When set, it means the dev_addr is a based on a value in the gid
table. this bit will be cleared upon a dev_addr change triggered
by the user and set after validation.

Per IB spec the port GUID can't change if the module is loaded.
port GUID is the basis for GID at index 0 which is the basis for
the default device address of a ipoib interface.

The issue is that there are devices that don't follow the spec,
they change the port GUID while HCA is powered on, so in order
not to break userspace applications. We need to check if the
user wanted to control the device address and we assume that
if he sets the device address back to be based on GID index 0,
he no longer wishs to control it.

In order to track this, we add an additional flag:
* IPOIB_FLAG_DEV_ADDR_CTRL

When setting the device address, there is no validation of the upper
twelve bytes of the device address (flags, qpn, subnet prefix) as those
bytes are not under the control of the user.

Signed-off-by: Mark Bloch 
Reviewed-by: Leon Romanovsky 
Signed-off-by: Leon Romanovsky 
Signed-off-by: Doug Ledford