Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

This patch makes following fixes in QP flush logic:

- correctly flushes unsignaled WRs followed by a signaled WR
- supports for flushing a CQ bound to multiple QPs
- resets cidx_flush if a active queue starts getting HW CQEs again
- marks WQ in error when we leave RTS. This was only being done for
  user queues, but we need it for kernel queues too so that
  post_send/post_recv will start returning the appropriate error
  synchronously
- eats unsignaled read resp CQEs. HW always inserts CQEs so we must
  silently discard them if the read work request was unsignaled.
- handles QP flushes with pending SW CQEs. The flush and out of order
  completion logic has a bug where if out of order completions are
  flushed but not yet polled by the consumer and the qp is then
  flushed then we end up inserting duplicate completions.
- c4iw_flush_sq() should only flush wrs that have not already been
  flushed.  Since we already track where in the SQ we've flushed via
  sq.cidx_flush, just start at that point and flush any remaining.
  This bug only caused a problem in the presence of unsignaled work
  requests.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NVipul Pandya <vipul@chelsio.com>

[ Fixed sparse warning due to htonl/ntohl confusion.  - Roland ]
Signed-off-by: NRoland Dreier <roland@purestorage.com>

"uresp.ma_sync_key" doesn't get set on this path so we leak 8 bytes of data.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Commit c079c287 ("RDMA/cxgb4: Fix error handling in create_qp()")
broke SQ allocation.  Instead of falling back to host allocation when
on-chip allocation fails, it tries to allocate both.  And when it
does, and we try to free the address from the genpool using the host
address, we hit a BUG and the system crashes as below.

We create a new function that has the previous behavior and properly
propagate the error, as intended.

    kernel BUG at /usr/src/packages/BUILD/kernel-ppc64-3.0.68/linux-3.0/lib/genalloc.c:340!
    Oops: Exception in kernel mode, sig: 5 [#1]
    SMP NR_CPUS=1024 NUMA pSeries
    Modules linked in: rdma_ucm rdma_cm ib_addr ib_cm iw_cm ib_sa ib_mad ib_uverbs iw_cxgb4 ib_core ip6t_LOG xt_tcpudp xt_pkttype ipt_LOG xt_limit ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_raw xt_NOTRACK ipt_REJECT xt_state iptable_raw iptable_filter ip6table_mangle nf_conntrack_netbios_ns nf_conntrack_broadcast nf_conntrack_ipv4 nf_conntrack nf_defrag_ipv4 ip_tables ip6table_filter ip6_tables x_tables fuse loop dm_mod ipv6 ipv6_lib sr_mod cdrom ibmveth(X) cxgb4 sg ext3 jbd mbcache sd_mod crc_t10dif scsi_dh_emc scsi_dh_hp_sw scsi_dh_alua scsi_dh_rdac scsi_dh ibmvscsic(X) scsi_transport_srp scsi_tgt scsi_mod
    Supported: Yes
    NIP: c00000000037d41c LR: d000000003913824 CTR: c00000000037d3b0
    REGS: c0000001f350ae50 TRAP: 0700   Tainted: G            X  (3.0.68-0.9-ppc64)
    MSR: 8000000000029032 <EE,ME,CE,IR,DR>  CR: 24042482  XER: 00000001
    TASK = c0000001f6f2a840[3616] 'rping' THREAD: c0000001f3508000 CPU: 0
    GPR00: c0000001f6e875c8 c0000001f350b0d0 c000000000fc9690 c0000001f6e875c0
    GPR04: 00000000000c0000 0000000000010000 0000000000000000 c0000000009d482a
    GPR08: 000000006a170000 0000000000100000 c0000001f350b140 c0000001f6e875c8
    GPR12: d000000003915dd0 c000000003f40000 000000003e3ecfa8 c0000001f350bea0
    GPR16: c0000001f350bcd0 00000000003c0000 0000000000040100 c0000001f6e74a80
    GPR20: d00000000399a898 c0000001f6e74ac8 c0000001fad91600 c0000001f6e74ab0
    GPR24: c0000001f7d23f80 0000000000000000 0000000000000002 000000006a170000
    GPR28: 000000000000000c c0000001f584c8d0 d000000003925180 c0000001f6e875c8
    NIP [c00000000037d41c] .gen_pool_free+0x6c/0xf8
    LR [d000000003913824] .c4iw_ocqp_pool_free+0x8c/0xd8 [iw_cxgb4]
    Call Trace:
    [c0000001f350b0d0] [c0000001f350b180] 0xc0000001f350b180 (unreliable)
    [c0000001f350b170] [d000000003913824] .c4iw_ocqp_pool_free+0x8c/0xd8 [iw_cxgb4]
    [c0000001f350b210] [d00000000390fd70] .dealloc_sq+0x90/0xb0 [iw_cxgb4]
    [c0000001f350b280] [d00000000390fe08] .destroy_qp+0x78/0xf8 [iw_cxgb4]
    [c0000001f350b310] [d000000003912738] .c4iw_destroy_qp+0x208/0x2d0 [iw_cxgb4]
    [c0000001f350b460] [d000000003861874] .ib_destroy_qp+0x5c/0x130 [ib_core]
    [c0000001f350b510] [d0000000039911bc] .ib_uverbs_cleanup_ucontext+0x174/0x4f8 [ib_uverbs]
    [c0000001f350b5f0] [d000000003991568] .ib_uverbs_close+0x28/0x70 [ib_uverbs]
    [c0000001f350b670] [c0000000001e7b2c] .__fput+0xdc/0x278
    [c0000001f350b720] [c0000000001a9590] .remove_vma+0x68/0xd8
    [c0000001f350b7b0] [c0000000001a9720] .exit_mmap+0x120/0x160
    [c0000001f350b8d0] [c0000000000af330] .mmput+0x80/0x160
    [c0000001f350b960] [c0000000000b5d0c] .exit_mm+0x1ac/0x1e8
    [c0000001f350ba10] [c0000000000b8154] .do_exit+0x1b4/0x4b8
    [c0000001f350bad0] [c0000000000b84b0] .do_group_exit+0x58/0xf8
    [c0000001f350bb60] [c0000000000ce9f4] .get_signal_to_deliver+0x2f4/0x5d0
    [c0000001f350bc60] [c000000000017ee4] .do_signal_pending+0x6c/0x3e0
    [c0000001f350bdb0] [c0000000000182cc] .do_signal+0x74/0x78
    [c0000001f350be30] [c000000000009e74] do_work+0x24/0x28
Signed-off-by: NThadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Cc: Emil Goode <emilgoode@gmail.com>
Cc: <stable@vger.kernel.org>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Fix to return a negative error code from the error handling case
instead of 0, as returned elsewhere in this function.
Signed-off-by: NWei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

T5 adapter does not support onchip queue memory. Present logic fails to
allocate QP for T5 and returns an error. Also, if module parameter ocqp_support
is zero then we are unable to allocate QP which should not be the case. Ideally
if ocqp_support parameter is 0 or onchip queue support is disable then host QP
should be allocated before returning an error.
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It enables direct DMA by HW to memory region PBL arrays and fast register PBL
arrays from host memory, vs the T4 way of passing these arrays in the WR itself.
The result is lower latency for memory registration, and larger PBL array
support for fast register operations.

This patch also updates ULP_TX_MEM_WRITE command fields for T5. Ordering bit of
ULP_TX_MEM_WRITE is at bit position 22 in T5 and at 23 in T4.
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Both DB Flow-Control and DB Coalescing are disabled by default on T5
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adds support for Chelsio T5 adapter.
Enables T5's Write Combining feature.
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If a FINI operation fails, then we need to ABORT instead of CLOSE.
Also, if we ABORT due to unexpected STREAMING data, then wake up
anybody blocked in FINI...
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The variable ret is assigned return values in a couple of places, but
its value is never returned.  This patch makes use of the ret variable
so that the caller get correct error codes returned.

The following changes are also introduced:

- The alloc_oc_sq function can return -ENOSYS or -ENOMEM so we want to
  get the return value from it.

- Change the label names to improve readability.
Signed-off-by: NEmil Goode <emilgoode@gmail.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

cxgb4 driver has duplicate definitions of registers which will be removed. This
patch updates the RDMA/cxgb4 driver accordingly.
Signed-off-by: NSantosh Rastapur <santosh@chelsio.com>
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Reviewed-by: NSivakumar Subramani <sivasu@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This allows querying the QP state before flushing.
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Add module option db_fc_threshold which is the count of active QPs
that trigger automatic db flow control mode.  Automatically transition
to/from flow control mode when the active qp count crosses
db_fc_theshold.

Add more db debugfs stats

On DB DROP event from the LLD, recover all the iwarp queues.
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Get FULL/EMPTY/DROP events from LLD.  On FULL event, disable normal
user mode DB rings.

Add modify_qp semantics to allow user processes to call into the
kernel to ring doobells without overflowing.

Add DB Full/Empty/Drop stats.

Mark queues when created indicating the doorbell state.

If we're in the middle of db overflow avoidance, then newly created
queues should start out in this mode.

Bump the C4IW_UVERBS_ABI_VERSION to 2 so the user mode library can
know if the driver supports the kernel mode db ringing.
Signed-off-by: NVipul Pandya <vipul@chelsio.com>
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

They had been getting it implicitly via device.h but we can't
rely on that for the future, due to a pending cleanup so fix
it now.
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>

QPs need to be moved to error before telling the firwmare to shutdown
the queue.  Otherwise, the application can submit WRs that will never
get fetched by the hardware and never flushed by the driver.
Signed-off-by: NKumar Sanghvi <kumaras@chelsio.com>
Acked-by: NSteve Wise <swsie@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Commit 01e7da6b ("RDMA/cxgb4: Make sure flush CQ entries are
collected on connection close") introduced a potential problem where a
CQ's comp_handler can get called simultaneously from different places
in the iw_cxgb4 driver.  This does not comply with
Documentation/infiniband/core_locking.txt, which states that at a
given point of time, there should be only one callback per CQ should
be active.

This problem was reported by Parav Pandit <Parav.Pandit@Emulex.Com>.
Based on discussion between Parav Pandit and Steve Wise, this patch
fixes the above problem by serializing the calls to a CQ's
comp_handler using a spin_lock.
Reported-by: NParav Pandit <Parav.Pandit@Emulex.Com>
Signed-off-by: NKumar Sanghvi <kumaras@chelsio.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

At the time when a peer closes the connection, iw_cxgb4 will not send
a cq event if ibqp.uobject exists.  In that case, its possible for a
user application to get blocked in ibv_get_cq_event().

To resolve this, call the cq's comp_handler to unblock any read from
ibv_get_cq_event().  This will trigger userspace to poll the cq and
collect flush status completions for any pending work requests.
Signed-off-by: NKumar Sanghvi <kumaras@chelsio.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

This patch adds support for Enhanced RDMA Connection Establishment
(draft-ietf-storm-mpa-peer-connect-06), aka MPAv2.  Details of draft
can be obtained from:
<http://www.ietf.org/id/draft-ietf-storm-mpa-peer-connect-06.txt>

The patch updates the following functions for initiator perspective:
 - send_mpa_request
 - process_mpa_reply
 - post_terminate for TERM error codes
 - destroy_qp for TERM related change
 - adds layer/etype/ecode to c4iw_qp_attrs for sending with TERM
 - peer_abort for retrying connection attempt with MPA_v1 message
 - added c4iw_reconnect function

The patch updates the following functions for responder perspective:
 - process_mpa_request
 - send_mpa_reply
 - c4iw_accept_cr
 - passes ird/ord to upper layers
Signed-off-by: NKumar Sanghvi <kumaras@chelsio.com>
Reviewed-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

- fix a race where the driver could end up sending a close_con_req
  after an abort_rpl.  In c4iw_ep_disconnect(), send abort or close
  request with the ep mutex held.

- fix a hang where driver fails to wake up when a connection is reset
  during a normal close.  Wake up any waiters in the interrupt path,
  and correctly cleanup after rdma_fini() failures.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Parens are missing: '|' has a higher presedence than '?'.
Signed-off-by: NRoel Kluin <roel.kluin@gmail.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Concurrent ingress CLOSE and ULP ABORT operations causes a crash due
to a race condition where the close path releases the EP lock and then
tries to move the QP state to CLOSED.  This must be done inside the EP
lock to avoid the race.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

MAXBURST cannot exceed 256B for on-chip queues.  With a 512B MAXBURST,
we can lock up the chip.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Cc: <stable@kernel.org>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Re-initializing the wait object in rdma_init()/rdma_fini() causes a
timing window which can lead to a deadlock during close.  Once this
deadlock hits, all RDMA activity over the T4 device will be stuck.

There's no need to re-init the wait object, so remove it.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Cc: <stable@kernel.org>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

This removes unused code found by running 'make namespacecheck';
compile tested only.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

- Remove dsgl support - doesn't work in T4.
- Wrap the immediate PBL as needed when building it in the wr.
- Adjust max pbl depth allowed based on ulptx alignment requirements.
- Bump the slots per SQ to 5 to allow up to 128MB fast registers.
- Advertise fastreg support by default.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

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

Move the connection setup/teardown paths to the workq thread removing
spin lock/irq disable requirements for these paths.  This allows calls
down to the LLD for EP and QP state transition actions to be atomic
with respect to processing CPL messages coming up from the HW.
Namely, calls to rdma_init() and rdma_fini() can now be called with
the mutex held avoiding many race conditions with the abort path.

The QP spinlock is still used but only to manipulate the qp state.  This
allows the fastpaths, poll, post_send, and pos_recv, to run in the
irq context.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

T4 support on-chip SQs to reduce latency.  This patch adds support for
this in iw_cxgb4:

 - Manage ocqp memory like other adapter mem resources.
 - Allocate user mode SQs from ocqp mem if available.
 - Map ocqp mem to user process using write combining.
 - Map PCIE_MA_SYNC reg to user process.

Bump uverbs ABI.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

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

T4 FW sends up CPL_RDMA_TERMINATE to indicate a peer TERM.  This
triggers the QP moving to TERMINATE state.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The HW design requires zeroing any pad in SGLs.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

In c4iw_modify_qp() error path, only use qhp->ep if ep is not already set.
Otherwise qhp->ep can be NULL and we crash.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix:

  drivers/infiniband/hw/cxgb4/qp.c: In function ‘create_qp’:
  drivers/infiniband/hw/cxgb4/qp.c:147: warning: cast from pointer to integer of different size
  drivers/infiniband/hw/cxgb4/qp.c: In function ‘rdma_fini’:
  drivers/infiniband/hw/cxgb4/qp.c:988: warning: cast from pointer to integer of different size
  drivers/infiniband/hw/cxgb4/qp.c: In function ‘rdma_init’:
  drivers/infiniband/hw/cxgb4/qp.c:1063: warning: cast from pointer to integer of different size
  drivers/infiniband/hw/cxgb4/mem.c: In function ‘write_adapter_mem’:
  drivers/infiniband/hw/cxgb4/mem.c:74: warning: cast from pointer to integer of different size
  drivers/infiniband/hw/cxgb4/cq.c: In function ‘destroy_cq’:
  drivers/infiniband/hw/cxgb4/cq.c:58: warning: cast from pointer to integer of different size
  drivers/infiniband/hw/cxgb4/cq.c: In function ‘create_cq’:
  drivers/infiniband/hw/cxgb4/cq.c:135: warning: cast from pointer to integer of different size
  drivers/infiniband/hw/cxgb4/cm.c: In function ‘fw6_msg’:
  drivers/infiniband/hw/cxgb4/cm.c:2326: warning: cast to pointer from integer of different size

by casting pointers to unsigned long instead of u64.
Reported-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

There exists a race condition where the app disconnects, which
initiates an orderly close (via rdma_fini()), concurrently with an
ingress abort condition, which initiates an abortive close operation.
Since rdma_fini() must be called without IRQs disabled, the fini can
be called after the QP has been transitioned to ERROR.  This is ok,
but we need to protect against qp->ep getting NULLed.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

T4 EQ entries are in multiples of 64 bytes.  Currently the RDMA SQ and
RQ use fixed sized entries composed of 4 EQ entries for the SQ and 2
EQ entries for the RQ.  For optimial latency with small IO, we need to
change this so the HW only needs to DMA the EQ entries actually used
by a given work request.

Implementation:

- add wq_pidx counter to track where we are in the EQ.  cidx/pidx are
  used for the sw sq/rq tracking and flow control.

- the variable part of work requests is the SGL.  Add new functions to
  build the SGL and/or immediate data directly in the EQ memory
  wrapping when needed.

- adjust the min burst size for the EQ contexts to 64B.
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The alloc_skb() in various allocations are failable, so remove
__GFP_NOFAIL from their masks.
Signed-off-by: NDavid Rientjes <rientjes@google.com>
Acked-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>