Use new ib_init_ah_from_wc() and ib_init_ah_from_path() helper
functions to clean up the IB CM.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add a call to initialize address handle attributes given a path record.
This is used by the CM, and would be useful for users of UD QPs.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add a function to initialize address handle attributes from a work
completion.  This functionality is duplicated by both verbs and the CM.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The P_Key is provided into a SIDR REQ in two places, once as a
parameter, and again in the path record.  Remove the P_Key as a
parameter and always use the one given in the path record.

This change has no practical effect on ABI functionality.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Misc cleanups in ib_srp:
1) I think that it is more efficient to move the req entries from req_list
   to free_list in srp_reconnect_target (rather than rebuild the free_list).
   (In any case this code is shorter).
2) This allows us to reuse code in srp_reset_device and srp_reconnect_target
   and call a new function srp_reset_req.
Signed-off-by: NIshai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

There has been a change in the format of port identifiers between
revision 10 of the SRP specification and the current revision 16A.

Revision 10 specifies port identifier format as

  lower 8 bytes :  GUID   upper 8 bytes :  Extension

Whereas revision 16A specifies it as 

 lower 8 bytes :  Extension  upper 8 bytes :  GUID

There are older targets (e.g. SilverStorm Virtual Fibre Channel
Bridge) which conform to revision 10 of the SRP specification.

The I/O class of revision 10 is 0xFF00 and the I/O class of revision
16A is 0x0100.

For supporting older targets, this patch:

1) Adds a new optional target creation parameter "io_class". Default
   value of io_class is 0x0100 (i.e. revision 16A)
2) Uses the correct port identifier format for targets with IO class
   of 0xFF00 (i.e. conforming to revision 10)
Signed-off-by: NRamachandra K <rkuchimanchi@silverstorm.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

When creating a FMR pool, query the IB device and use the returned
max_map_map_per_fmr attribute as for the max number of FMR remaps. If
the device does not suport querying this attribute, use the original
IB_FMR_MAX_REMAPS (32) default.
Signed-off-by: NOr Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Report the true max_map_per_fmr value from mthca_query_device(),
taking into account the change in FMR remapping introduced by the
Sinai performance optimization.
Signed-off-by: NOr Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Generate a client reregister event instead of a LID change event when
client reregister bit is set.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Change the mthca snoop of MADs that set PortInfo to check if the SM
has set the client reregister bit, and if it has, generate a client
reregister event.  If the bit is not set, just generate a LID change
event as usual.
Signed-off-by: NLeonid Arsh <leonida@voltaire.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Move ipath's struct port_info into <rdma/ib_smi.h>, so that it can be
used by mthca to implement client reregister support.

Remove the __attribute__((packed)) because all the members of the struct
are naturally aligned anyway.
Signed-off-by: NLeonid Arsh <leonida@voltaire.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Handle client reregister events by treating them just like LID or
SM changes -- flush all cached paths and rejoin multicast groups.
Signed-off-by: NLeonid Arsh <leonida@voltaire.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix misaligned access faults on ia64: never cast a misaligned
neighbour->ha + 4 pointer to union ib_gid type; pass a void * pointer
instead.  The memcpy was being optimized to use full word accesses
because the compiler thought that union ib_gid is always aligned.

The cast in IPOIB_GID_ARG is safe, since it is fixed to access each
byte separately.
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The comparisons of priv->tx_tail to ah->last_send in ipoib_free_ah()
and ipoib_post_receive() are slightly unsafe, because priv->tx_lock is
not held and hence a stale value of ah->last_send might be used, which
would lead to freeing an AH before the driver was really done with it.
The simple way to fix this is to the optimization of early free from
ipoib_free_ah() and unconditionally queue AHs for reaping, and then
take priv->tx_lock in __ipoib_reap_ah().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Check GID/LID for requester side when searching for request which
matches received response.  This is in order to guarantee uniqueness
if the same TID is used when requesting via multiple source LIDs (when
LMC is not zero).  Use ports' cached LMC to perform the check.

Further, do not perform LID check for direct-routed packets, since
the permissive LID makes a proper check impossible.
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add an LMC cache to struct ib_device, and add a function
ib_get_cached_lmc() to query the cache.
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

destroy_workqueue() already does flush_workqueue().
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NSean Hefty <sean.hefty@intel.com>

Convert semaphore in ib_ucm_file to a real mutex.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

It's perfectly valid for a connection to an SRP target to have a
request limit of 0, so get rid of the message about it, which can spam
kernel logs even with printk_ratelimit().  Keep a count of such events
in a "zero_req_lim" SCSI host attribute instead, so someone who cares
can look at the statistics.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Handle IB_CM_DREQ_ERROR and IB_CM_DREQ_RECEIVED events from the CM,
instead of just printing "Unhandled CM event".  In the case of
DREQ_ERROR, just ignore the event -- a TIMEWAIT_EXIT will be generated
also.  For DREQ_RECEIVED, send a DREP in response to shut the
connection down cleanly.
Signed-off-by: NIshai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Make the sg_tablesize used by SRP adjustable at module load time via a
module parameter.  Calculate the corresponding IU length required to
support this.
Signed-off-by: NVu Pham <vu@mellanox.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Allow userspace to throttle traffic on a given connection to a target
port by adding "max_cmd_per_lun=xyz" to lower the cmd_per_lun value
set for that scsi_host.
Signed-off-by: NVu Pham <vu@mellanox.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Interrupts will always be enabled in srp_remove_one(), so
spin_lock_irq() can be used instead of spin_lock_irqsave().
Also, the loop takes target->scsi_host->host_lock, so target->state
can just be set to SRP_TARGET_REMOVED witout testing the old value.
Signed-off-by: NIshai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The SRP driver never sleeps while holding target_mutex, and it's just
used to protect some simple list operations, so hold times will be
short.  So just convert it to a spinlock, which is smaller and faster.
Signed-off-by: NMatthew Wilcox <matthew@wil.cx>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

list_for_each_entry_safe() is used in one place where the list isn't
modified.  So just change it to list_for_each_entry().
Signed-off-by: NMatthew Wilcox <matthew@wil.cx>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

SCAN_WILD_CARD is indeed available from <scsi/scsi.h>, which is
already included.  So get rid of private hack.
Signed-off-by: NMatthew Wilcox <matthew@wil.cx>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The kernel has had wait_for_completion_timeout() for a long time now.
mthca should use it to handle FW commands timing out, instead of
implementing the same thing in a much more complicated way by using
wait_for_completion() along with a timer that does complete().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Create an SRP FMR pool on HCAs that support FMRs, and use FMRs to map
gather/scatter lists that have more than one entry into a single
memory region that appears virtually contiguous to the SRP target
(which is the RDMA initiator).

This patch bails out on FMR mapping for SCSI commands where the
gather/scatter list cannot be mapped into a single FMR because there
are sub-page-sized entries in middle of the list.  An unaligned
start or end of the list is OK.

Based on a patch by Vu Pham <vuhuong@mellanox.com>.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Kill some dead code in mthca_eq.c
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Kernel connection management agent over InfiniBand that connects based
on IP addresses.  The agent defines a generic RDMA connection
abstraction to support clients wanting to connect over different RDMA
devices.

The agent also handles RDMA device hotplug events on behalf of clients.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add an address translation service that maps IP addresses to
InfiniBand GID addresses using IPoIB.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Extend matching connection requests to listens in the InfiniBand CM to
include private data checks.

This allows applications to listen on the same service identifier,
with private data directing the request to the appropriate application.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Provide common handling for marshalling data between userspace clients
and kernel InfiniBand drivers.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Memfree firmware is in rare cases reporting WQE index == base - 1 in
receive completion with error, instead of (rq size - 1); base is 0 in
mthca.  Here is a patch to avoid kernel crash and report a correct WR
id in this case.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

mthca does not restore the following PCI-X/PCI Express registers after reset:
  PCI-X device: PCI-X command register
  PCI-X bridge: upstream and downstream split transaction registers
  PCI Express : PCI Express device control and link control registers

This causes instability and/or bad performance on systems where one of
these registers is set to a non-default value by BIOS.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

When ipoib_stop() is called it first calls netif_stop_queue() to stop
the kernel from passing more packets to the network driver. However,
the completion handler may call netif_wake_queue() re-enabling packet
transfer.

This might result in leaks (we see AH leaks which we think can be
attributed to this bug) as new packets get posted while the interface
is going down.
Signed-off-by: NEli Cohen <eli@mellanox.co.il>
Signed-off-by: NMichael Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

If we post a list of length exactly a multiple of 256, nreq in
doorbell gets set to 256 which is wrong: it should be encoded by 0.
This is because we only zero it out on the next WR, which may not be
there.  The solution is to ring the doorbell after posting a WQE, not
before posting the next one.

This is the same bug that we just fixed for QPs with non-shared RQ.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

At this point, the core QP structure hasn't been initialized, so what's
in there isn't valid.  Get the same information elsewhere.
Signed-off-by: NBryan O'Sullivan <bos@pathscale.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The problem was that node A's sending thread, which handles sending RDMA
read response data, would write the trigger word, the last packet would
be sent, node B would send a new RDMA read request, node A's interrupt
handler would initialize s_rdma_sge, then node A's sending thread would
update s_rdma_sge.  This didn't happen very often naturally but was more
frequent with 1 byte RDMA reads.  Rather than adding more locking or
increasing the QP structure size and copying sge data, I modified the
copy routine to update the pointers before writing the trigger word to
avoid the update race.
Signed-off-by: NRalph Campbell <ralphc@pathscale.com>
Signed-off-by: NBryan O'Sullivan <bos@pathscale.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>