Conflicts:
	drivers/infiniband/core/uverbs_main.c

Instead of repeating the error unwinding steps in each place an error
can be detected, use the common idiom of gotos into an error flow.
Signed-off-by: Roland Dreier <rolandd@cisco.com>

We can reduce the number of IB interrupts from two interrupts per
srp_queuecommand() call to one by using separate CQs for send and
receive completions and processing send completions by polling every
time a TX IU is allocated.

Receive completion events still trigger an interrupt.
Signed-off-by: Bart Van Assche <bart.vanassche@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Add support for KR device id 0x0110.  While at it, cleanup
nes_init_phy() by splitting it into nes_init_1g_phy() and
nes_init_2025_phy().

Remove support for NES_PHY_TYPE_IRIS, which was used on an XFP board
that was only manufactured in small quantities and given out for evals
in even smaller quantities.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

The anon_inodes interface has been split to allow creating a bare
(non-installed) file pointer and also extended to allow specifying
O_RDONLY in the flags.  This makes it a suitable replacement for the
private "infinibandeventfs" pseudo-filesystem used by uverbs, and this
replacement saves a small chunk of boilerplate code.
Signed-off-by: Roland Dreier <rolandd@cisco.com>

ib_ud_header_init() first clears header and then fills up the various
fields.  Later on, it tests header->immediate_present, which it has
already cleared, so the condition is always false.  Fix this by adding
an immediate_present parameter and setting header->immediate_present
as is done with grh_present.  Also remove unused calculation of
header_len.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

If cxgb3 calls the iw_cxgb3 t3cclient remove function due to a device
removal event, then the iwch device must be marked with CXIO_ERROR_FATAL
since the device below us is going away.  Otherwise, we can get stuck in
a deadlock as RDMA ULPs try and deallocate objects (like MRs, QPs, etc).
So always mark the device with CXIO_ERROR_FATAL when removing.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Only kernel mode CQs need the SW queue memory allocated.  The SW queue
for user mode CQs is allocated in userspace by libcxgb3.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

T3 hardware doorbell FIFO overflows can cause application stalls due
to lost doorbell ring events.  This has been seen when running large
NP IMB alltoall MPI jobs.  The T3 hardware supports an xon/xoff-type
flow control mechanism to help avoid overflowing the HW doorbell FIFO.

This patch uses these interrupts to disable RDMA QP doorbell rings
when we near an overflow condition, and then turn them back on (and
ring all the active QP doorbells) when when the doorbell FIFO empties
out.  In addition if an doorbell ring is dropped by the hardware, the
code will now recover.

Design:

cxgb3:
- enable these DB interrupts
- in the interrupt handler, schedule work tasks to call the ULPs event
  handlers with the new events.
- ring all the qset txqs when an overflow is detected.

iw_cxgb3:
- disable db ringing on all active qps when we get the DB_FULL event
- enable db ringing on all active qps and ring all active dbs when we get
  the DB_EMPTY event
- On DB_DROP event:
       - disable db rings in the event handler
       - delay-schedule a work task which rings and enables the dbs on
         all active qps.
- in post_send and post_recv logic, don't ring the db if it's disabled.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

A small change to reduce the size of ib_device to 1112 bytes
(from 1128).
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

As shown when 'let c_space_errors=1' is set in vim.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Some large systems may support more than IB_UCM_MAX_DEVICES
(currently 32).

This change allows us to support more devices in a backwards-compatible
manner. the first IB_UCM_MAX_DEVICES keep the same major/minor device
numbers they've always had.

If there are more than IB_UCM_MAX_DEVICES, then we dynamically request
a new major device number (new minors start at 0).
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

This change is not useful by itself, but sets us up for a future
change that allows us to support more than IB_UCM_MAX_DEVICES.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

This change is not useful by itself, but sets us up for a future
change that allows us to dynamically allocate device numbers in case
we have more than IB_UCM_MAX_DEVICES in the system.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Clean errors as shown when 'let c_space_errors=1' is set in vim.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Some large systems may support more than IB_UMAD_MAX_PORTS
(currently 64).

This change allows us to support more ports in a backwards-compatible
manner.  The first IB_UMAD_MAX_PORTS keep the same major/minor device
numbers they've always had.

If there are more than IB_UMAD_MAX_PORTS, we then dynamically request
a new major device number (new minors start at 0).
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

This change is not useful by itself, but sets us up for a future change
that allows us to support more than IB_UMAD_MAX_PORTS in a system.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

This change is not useful by itself, but sets us up for a future
change that allows us to dynamically allocate device numbers in case
we have more than IB_UMAD_MAX_PORTS in the system.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

We no longer need this data structure, as it was used to associate an
inode back to a struct ib_umad_port during ->open().  But now that
we're embedding a struct cdev in struct ib_umad_port, we can use the
container_of() macro to go from the inode back to the device instead.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Instead of storing pointers to cdev and sm_cdev, embed the full
structures instead.

This change allows us to use the container_of() macro in ib_umad_open()
and ib_umad_sm_open() in a future patch.

This change increases the size of struct ib_umad_port to 320 bytes
from 128.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Clean up the errors as shown when 'let c_space_errors=1' is set in vim.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Eliminate some padding in the structure by rearranging the members.
sizeof(struct ib_uverbs_event_file) is now 72 bytes (from 80) and
more members now fit in the first cacheline.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Some large systems may support more than IB_UVERBS_MAX_DEVICES
(currently 32).

This change allows us to support more devices in a backwards-compatible
manner.  The first IB_UVERBS_MAX_DEVICES keep the same major/minor
device numbers that they've always had.

If there are more than IB_UVERBS_MAX_DEVICES, we then dynamically
request a new major device number (new minors start at 0).

This change increases the maximum number of HCAs to 64 (from 32).
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

This change is not useful by itself, but sets us up for a future change
that allows us to support more than IB_UVERBS_MAX_DEVICES in a system.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

This change is not useful by itself, but it sets us up for a future
change that allows us to dynamically allocate device numbers in case
we have more than IB_UVERBS_MAX_DEVICES in the system.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

dev_table's raison d'etre was to associate an inode back to a struct
ib_uverbs_device.

However, now that we've converted ib_uverbs_device to contain an
embedded cdev (instead of a *cdev), we can use the container_of()
macro and cast back to the containing device.

There's no longer any need for dev_table, so get rid of it.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Instead of storing a pointer to a cdev, embed the entire struct cdev.

This change allows us to use the container_of() macro in
ib_uverbs_open() in a future patch.

This change increases the size of struct ib_uverbs_device to 168 bytes
across 3 cachelines from 80 bytes in 2 cachelines.  However, we
rearrange the members so that everything fits into the first cacheline
except for the struct cdev. Finally, we don't touch the cdev in any
fastpaths, so this change shouldn't negatively affect performance.
Signed-off-by: Alex Chiang <achiang@hp.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Currently the iSER receive completion flow takes the session lock
twice.  Optimize it to avoid the first one by letting
iser_task_rdma_finalize() be called only from the cleanup_task
callback invoked by iscsi_free_task, thus reducing the contention on
the session lock between the scsi command submission to the scsi
command completion flows.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

libiscsi passthrough mode invokes the transport xmit calls directly
without first going through an internal queue, unlike the other mode,
which uses a queue and a xmitworker thread.  Now that the "cant_sleep"
prerequisite of iscsi_host_alloc is met, move to use it.  Handling
xmit errors is now done by the passthrough flow of libiscsi.  Since
the queue/worker aren't used in this mode, the code that schedules the
xmitworker is removed.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Remove unnecessary checks for the IB connection state and for QP
overflow, as conn state changes are reported by iSER to libiscsi and
handled there. QP overflow is theoretically possible only when
unsolicited data-outs are used; anyway it's being checked and handled
by HW drivers.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

Two minor flows in iSER's data path still use allocations; move them
to be atomic as a preperation step towards moving to use libiscsi
passthrough mode.
Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>