It's not safe to use resources pointed to by the @send_wr of
ib_post_send() _after_ that function returns. Those resources are
typically freed by the Send completion handler, which can run before
ib_post_send() returns.

Thus the trace points currently around ib_post_send() in the
client's RPC/RDMA transport are a hazard, even when they are
disabled. Rearrange them so that they touch the Work Request only
_before_ ib_post_send() is invoked.

Fixes: ab03eff5 ("xprtrdma: Add trace points in RPC Call transmit paths")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Commit e28ce900 ("xprtrdma: kmalloc rpcrdma_ep separate from
rpcrdma_xprt") erroneously removed a xprt_force_disconnect()
call from the "transport disconnect" path. The result was that the
client no longer responded to server-side disconnect requests.

Restore that call.

Fixes: e28ce900 ("xprtrdma: kmalloc rpcrdma_ep separate from rpcrdma_xprt")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Change the rpcrdma_xprt_disconnect() function so that it no longer
waits for the DISCONNECTED event.  This prevents blocking if the
remote is unresponsive.

In rpcrdma_xprt_disconnect(), the transport's rpcrdma_ep is
detached. Upon return from rpcrdma_xprt_disconnect(), the transport
(r_xprt) is ready immediately for a new connection.

The RDMA_CM_DEVICE_REMOVAL and RDMA_CM_DISCONNECTED events are now
handled almost identically.

However, because the lifetimes of rpcrdma_xprt structures and
rpcrdma_ep structures are now independent, creating an rpcrdma_ep
needs to take a module ref count. The ep now owns most of the
hardware resources for a transport.

Also, a kref is needed to ensure that rpcrdma_ep sticks around
long enough for the cm_event_handler to finish.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

rpcrdma_cm_event_handler() is always passed an @id pointer that is
valid. However, in a subsequent patch, we won't be able to extract
an r_xprt in every case. So instead of using the r_xprt's
presentation address strings, extract them from struct rdma_cm_id.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

I eventually want to allocate rpcrdma_ep separately from struct
rpcrdma_xprt so that on occasion there can be more than one ep per
xprt.

The new struct rpcrdma_ep will contain all the fields currently in
rpcrdma_ia and in rpcrdma_ep. This is all the device and CM settings
for the connection, in addition to per-connection settings
negotiated with the remote.

Take this opportunity to rename the existing ep fields from rep_* to
re_* to disambiguate these from struct rpcrdma_rep.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Completion errors after a disconnect often occur much sooner than a
CM_DISCONNECT event. Use this to try to detect connection loss more
quickly.

Note that other kernel ULPs do take care to disconnect explicitly
when a WR is flushed.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up:
The upper layer serializes calls to xprt_rdma_close, so there is no
need for an atomic bit operation, saving 8 bytes in rpcrdma_ia.

This enables merging rpcrdma_ia_remove directly into the disconnect
logic.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Move rdma_cm_id creation into rpcrdma_ep_create() so that it is now
responsible for allocating all per-connection hardware resources.

With this clean-up, all three arms of the switch statement in
rpcrdma_ep_connect are exactly the same now, thus the switch can be
removed.

Because device removal behaves a little differently than
disconnection, there is a little more work to be done before
rpcrdma_ep_destroy() can release the connection's rdma_cm_id. So
it is not quite symmetrical with rpcrdma_ep_create() yet.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Make a Protection Domain (PD) a per-connection resource rather than
a per-transport resource. In other words, when the connection
terminates, the PD is destroyed.

Thus there is one less HW resource that remains allocated to a
transport after a connection is closed.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Simplify the synopses of functions in the connect and
disconnect paths in preparation for combining the rpcrdma_ia and
struct rpcrdma_ep structures.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Simplify the synopses of functions in the post_send path
by combining the struct rpcrdma_ia and struct rpcrdma_ep arguments.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: prepare for combining the rpcrdma_ia and rpcrdma_ep
structures. Take the opportunity to rename the function to be
consistent with the "subsystem _ object _ verb" naming scheme.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Refactor rpcrdma_ep_create(), rpcrdma_ep_disconnect(), and
rpcrdma_ep_destroy().

rpcrdma_ep_create will be invoked at connect time instead of at
transport set-up time. It will be responsible for allocating per-
connection resources. In this patch it allocates the CQs and
creates a QP. More to come.

rpcrdma_ep_destroy() is the inverse functionality that is
invoked at disconnect time. It will be responsible for releasing
the CQs and QP.

These changes should be safe to do because both connect and
disconnect is guaranteed to be serialized by the transport send
lock.

This takes us another step closer to resolving the address and route
only at connect time so that connection failover to another device
will work correctly.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: This simplifies the logic in rpcrdma_post_recvs.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

To safely get rid of all rpcrdma_reps from a particular connection
instance, xprtrdma has to wait until each of those reps is finished
being used. A rep may be backing the rq_rcv_buf of an RPC that has
just completed, for example.

Since it is safe to invoke rpcrdma_rep_destroy() only in the Receive
completion handler, simply mark reps remaining in the rb_all_reps
list after the transport is drained. These will then be deleted as
rpcrdma_post_recvs pulls them off the rep free list.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

This reduces the hardware and memory footprint of an unconnected
transport.

At some point in the future, transport reconnect will allow
resolving the destination IP address through a different device. The
current change enables reps for the new connection to be allocated
on whichever NUMA node the new device affines to after a reconnect.

Note that this does not destroy _all_ the transport's reps... there
will be a few that are still part of a running RPC completion.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Currently the underlying RDMA device is chosen at transport set-up
time. But it will soon be at connect time instead.

The maximum size of a transport header is based on device
capabilities. Thus transport header buffers have to be allocated
_after_ the underlying device has been chosen (via address and route
resolution); ie, in the connect worker.

Thus, move the allocation of transport header buffers to the connect
worker, after the point at which the underlying RDMA device has been
chosen.

This also means the RDMA device is available to do a DMA mapping of
these buffers at connect time, instead of in the hot I/O path. Make
that optimization as well.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Refactor: Perform the "is supported" check in rpcrdma_ep_create()
instead of in rpcrdma_ia_open(). frwr_open() is where most of the
logic to query device attributes is already located.

The current code displays a redundant error message when the device
does not support FRWR. As an additional clean-up, this patch removes
the extra message.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

To support device hotplug and migrating a connection between devices
of different capabilities, we have to guarantee that all in-kernel
devices can support the same max NFS payload size (1 megabyte).

This means that possibly one or two in-tree devices are no longer
supported for NFS/RDMA because they cannot support 1MB rsize/wsize.
The only one I confirmed was cxgb3, but it has already been removed
from the kernel.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: there is no need to keep two copies of the same value.
Also, in subsequent patches, rpcrdma_ep_create() will be called in
the connect worker rather than at set-up time.

Minor fix: Initialize the transport's sendctx to the value based on
the capabilities of the underlying device, not the maximum setting.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The size of the sendctx queue depends on the value stored in
ia->ri_max_send_sges. This value is determined by querying the
underlying device.

Eventually, rpcrdma_ia_open() and rpcrdma_ep_create() will be called
in the connect worker rather than at transport set-up time. The
underlying device will not have been chosen device set-up time.

The sendctx queue will thus have to be created after the underlying
device has been chosen via address and route resolution; in other
words, in the connect worker.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean-up. The max_send_sge value also happens to be stored in
ep->rep_attr. Let's keep just a single copy.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Since v5.4, a device removal occasionally triggered this oops:

Dec  2 17:13:53 manet kernel: BUG: unable to handle page fault for address: 0000000c00000219
Dec  2 17:13:53 manet kernel: #PF: supervisor read access in kernel mode
Dec  2 17:13:53 manet kernel: #PF: error_code(0x0000) - not-present page
Dec  2 17:13:53 manet kernel: PGD 0 P4D 0
Dec  2 17:13:53 manet kernel: Oops: 0000 [#1] SMP
Dec  2 17:13:53 manet kernel: CPU: 2 PID: 468 Comm: kworker/2:1H Tainted: G        W         5.4.0-00050-g53717e43af61 #883
Dec  2 17:13:53 manet kernel: Hardware name: Supermicro SYS-6028R-T/X10DRi, BIOS 1.1a 10/16/2015
Dec  2 17:13:53 manet kernel: Workqueue: ib-comp-wq ib_cq_poll_work [ib_core]
Dec  2 17:13:53 manet kernel: RIP: 0010:rpcrdma_wc_receive+0x7c/0xf6 [rpcrdma]
Dec  2 17:13:53 manet kernel: Code: 6d 8b 43 14 89 c1 89 45 78 48 89 4d 40 8b 43 2c 89 45 14 8b 43 20 89 45 18 48 8b 45 20 8b 53 14 48 8b 30 48 8b 40 10 48 8b 38 <48> 8b 87 18 02 00 00 48 85 c0 75 18 48 8b 05 1e 24 c4 e1 48 85 c0
Dec  2 17:13:53 manet kernel: RSP: 0018:ffffc900035dfe00 EFLAGS: 00010246
Dec  2 17:13:53 manet kernel: RAX: ffff888467290000 RBX: ffff88846c638400 RCX: 0000000000000048
Dec  2 17:13:53 manet kernel: RDX: 0000000000000048 RSI: 00000000f942e000 RDI: 0000000c00000001
Dec  2 17:13:53 manet kernel: RBP: ffff888467611b00 R08: ffff888464e4a3c4 R09: 0000000000000000
Dec  2 17:13:53 manet kernel: R10: ffffc900035dfc88 R11: fefefefefefefeff R12: ffff888865af4428
Dec  2 17:13:53 manet kernel: R13: ffff888466023000 R14: ffff88846c63f000 R15: 0000000000000010
Dec  2 17:13:53 manet kernel: FS:  0000000000000000(0000) GS:ffff88846fa80000(0000) knlGS:0000000000000000
Dec  2 17:13:53 manet kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
Dec  2 17:13:53 manet kernel: CR2: 0000000c00000219 CR3: 0000000002009002 CR4: 00000000001606e0
Dec  2 17:13:53 manet kernel: Call Trace:
Dec  2 17:13:53 manet kernel: __ib_process_cq+0x5c/0x14e [ib_core]
Dec  2 17:13:53 manet kernel: ib_cq_poll_work+0x26/0x70 [ib_core]
Dec  2 17:13:53 manet kernel: process_one_work+0x19d/0x2cd
Dec  2 17:13:53 manet kernel: ? cancel_delayed_work_sync+0xf/0xf
Dec  2 17:13:53 manet kernel: worker_thread+0x1a6/0x25a
Dec  2 17:13:53 manet kernel: ? cancel_delayed_work_sync+0xf/0xf
Dec  2 17:13:53 manet kernel: kthread+0xf4/0xf9
Dec  2 17:13:53 manet kernel: ? kthread_queue_delayed_work+0x74/0x74
Dec  2 17:13:53 manet kernel: ret_from_fork+0x24/0x30

The proximal cause is that this rpcrdma_rep has a rr_rdmabuf that
is still pointing to the old ib_device, which has been freed. The
only way that is possible is if this rpcrdma_rep was not destroyed
by rpcrdma_ia_remove.

Debugging showed that was indeed the case: this rpcrdma_rep was
still in use by a completing RPC at the time of the device removal,
and thus wasn't on the rep free list. So, it was not found by
rpcrdma_reps_destroy().

The fix is to introduce a list of all rpcrdma_reps so that they all
can be found when a device is removed. That list is used to perform
only regbuf DMA unmapping, replacing that call to
rpcrdma_reps_destroy().

Meanwhile, to prevent corruption of this list, I've moved the
destruction of temp rpcrdma_rep objects to rpcrdma_post_recvs().
rpcrdma_xprt_drain() ensures that post_recvs (and thus rep_destroy) is
not invoked while rpcrdma_reps_unmap is walking rb_all_reps, thus
protecting the rb_all_reps list.

Fixes: b0b227f0 ("xprtrdma: Use an llist to manage free rpcrdma_reps")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

I've found that on occasion, "rmmod <dev>" will hang while if an NFS
is under load.

Ensure that ri_remove_done is initialized only just before the
transport is woken up to force a close. This avoids the completion
possibly getting initialized again while the CM event handler is
waiting for a wake-up.

Fixes: bebd0318 ("xprtrdma: Support unplugging an HCA from under an NFS mount")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

On device re-insertion, the RDMA device driver crashes trying to set
up a new QP:

Nov 27 16:32:06 manet kernel: BUG: kernel NULL pointer dereference, address: 00000000000001c0
Nov 27 16:32:06 manet kernel: #PF: supervisor write access in kernel mode
Nov 27 16:32:06 manet kernel: #PF: error_code(0x0002) - not-present page
Nov 27 16:32:06 manet kernel: PGD 0 P4D 0
Nov 27 16:32:06 manet kernel: Oops: 0002 [#1] SMP
Nov 27 16:32:06 manet kernel: CPU: 1 PID: 345 Comm: kworker/u28:0 Tainted: G        W         5.4.0 #852
Nov 27 16:32:06 manet kernel: Hardware name: Supermicro SYS-6028R-T/X10DRi, BIOS 1.1a 10/16/2015
Nov 27 16:32:06 manet kernel: Workqueue: xprtiod xprt_rdma_connect_worker [rpcrdma]
Nov 27 16:32:06 manet kernel: RIP: 0010:atomic_try_cmpxchg+0x2/0x12
Nov 27 16:32:06 manet kernel: Code: ff ff 48 8b 04 24 5a c3 c6 07 00 0f 1f 40 00 c3 31 c0 48 81 ff 08 09 68 81 72 0c 31 c0 48 81 ff 83 0c 68 81 0f 92 c0 c3 8b 06 <f0> 0f b1 17 0f 94 c2 84 d2 75 02 89 06 88 d0 c3 53 ba 01 00 00 00
Nov 27 16:32:06 manet kernel: RSP: 0018:ffffc900035abbf0 EFLAGS: 00010046
Nov 27 16:32:06 manet kernel: RAX: 0000000000000000 RBX: 00000000000001c0 RCX: 0000000000000000
Nov 27 16:32:06 manet kernel: RDX: 0000000000000001 RSI: ffffc900035abbfc RDI: 00000000000001c0
Nov 27 16:32:06 manet kernel: RBP: ffffc900035abde0 R08: 000000000000000e R09: ffffffffffffc000
Nov 27 16:32:06 manet kernel: R10: 0000000000000000 R11: 000000000002e800 R12: ffff88886169d9f8
Nov 27 16:32:06 manet kernel: R13: ffff88886169d9f4 R14: 0000000000000246 R15: 0000000000000000
Nov 27 16:32:06 manet kernel: FS:  0000000000000000(0000) GS:ffff88846fa40000(0000) knlGS:0000000000000000
Nov 27 16:32:06 manet kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
Nov 27 16:32:06 manet kernel: CR2: 00000000000001c0 CR3: 0000000002009006 CR4: 00000000001606e0
Nov 27 16:32:06 manet kernel: Call Trace:
Nov 27 16:32:06 manet kernel: do_raw_spin_lock+0x2f/0x5a
Nov 27 16:32:06 manet kernel: create_qp_common.isra.47+0x856/0xadf [mlx4_ib]
Nov 27 16:32:06 manet kernel: ? slab_post_alloc_hook.isra.60+0xa/0x1a
Nov 27 16:32:06 manet kernel: ? __kmalloc+0x125/0x139
Nov 27 16:32:06 manet kernel: mlx4_ib_create_qp+0x57f/0x972 [mlx4_ib]

The fix is to copy the qp_init_attr struct that was just created by
rpcrdma_ep_create() instead of using the one from the previous
connection instance.

Fixes: 98ef77d1 ("xprtrdma: Send Queue size grows after a reconnect")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Use a single trace point to record each connection's
negotiated inline thresholds and the computed maximum byte size
of transport headers.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

For debugging, the op_connect trace point should report the computed
connect delay. We can then ensure that the delay is computed at the
proper times, for example.

As a further clean-up, remove a few low-value "heartbeat" trace
points in the connect path.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

On some platforms, DMA mapping part of a page is more costly than
copying bytes. Restore the pull-up code and use that when we
think it's going to be faster. The heuristic for now is to pull-up
when the size of the RPC message body fits in the buffer underlying
the head iovec.

Indeed, not involving the I/O MMU can help the RPC/RDMA transport
scale better for tiny I/Os across more RDMA devices. This is because
interaction with the I/O MMU is eliminated, as is handling a Send
completion, for each of these small I/Os. Without the explicit
unmapping, the NIC no longer needs to do a costly internal TLB shoot
down for buffers that are just a handful of bytes.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: This field is not needed in the Send completion handler,
so it can be moved to struct rpcrdma_req to reduce the size of
struct rpcrdma_sendctx, and to reduce the amount of memory that
is sloshed between the sending process and the Send completion
process.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Micro-optimization: Save eight bytes in a frequently allocated
structure.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

MRs are now allocated on demand so we can safely throw them away on
disconnect. This way an idle transport can disconnect and it won't
pin hardware MR resources.

Two additional changes:

- Now that all MRs are destroyed on disconnect, there's no need to
  check during header marshaling if a req has MRs to recycle. Each
  req is sent only once per connection, and now rl_registered is
  guaranteed to be empty when rpcrdma_marshal_req is invoked.

- Because MRs are now destroyed in a WQ_MEM_RECLAIM context, they
  also must be allocated in a WQ_MEM_RECLAIM context. This reduces
  the likelihood that device driver memory allocation will trigger
  memory reclaim during NFS writeback.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Close some holes introduced by commit 6dc6ec9e ("xprtrdma: Cache
free MRs in each rpcrdma_req") that could result in list corruption.

In addition, the result that is tabulated in @count is no longer
used, so @count is removed.

Fixes: 6dc6ec9e ("xprtrdma: Cache free MRs in each rpcrdma_req")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

A recent clean up attempted to separate Receive handling and RPC
Reply processing, in the name of clean layering.

Unfortunately, we can't do this because the Receive Queue has to be
refilled _after_ the most recent credit update from the responder
is parsed from the transport header, but _before_ we wake up the
next RPC sender. That is right in the middle of
rpcrdma_reply_handler().

Usually this isn't a problem because current responder
implementations don't vary their credit grant. The one exception is
when a connection is established: the grant goes from one to a much
larger number on the first Receive. The requester MUST post enough
Receives right then so that any outstanding requests can be sent
without risking RNR and connection loss.

Fixes: 6ceea368 ("xprtrdma: Refactor Receive accounting")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up/code de-duplication.

Nit: RPC_CWNDSHIFT is incorrect as the initial value for xprt->cwnd.
This mistake does not appear to have operational consequences, since
the cwnd value is replaced with a valid value upon the first Receive
completion.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

This is because xprt_request_get_cong() is allowing more than one
RPC Call to be transmitted before the first Receive on the new
connection. The first Receive fills the Receive Queue based on the
server's credit grant. Before that Receive, there is only a single
Receive WR posted because the client doesn't know the server's
credit grant.

Solution is to clear rq_cong on all outstanding rpc_rqsts when the
the cwnd is reset. This is because an RPC/RDMA credit is good for
one connection instance only.

Fixes: 75891f50 ("SUNRPC: Support for congestion control ... ")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Eli Dorfman reports that after a series of idle disconnects, an
RPC/RDMA transport becomes unusable (rdma_create_qp returns
-ENOMEM). Problem was tracked down to increasing Send Queue size
after each reconnect.

The rdma_create_qp() API does not promise to leave its @qp_init_attr
parameter unaltered. In fact, some drivers do modify one or more of
its fields. Thus our calls to rdma_create_qp must use a fresh copy
of ib_qp_init_attr each time.

This fix is appropriate for kernels dating back to late 2007, though
it will have to be adapted, as the connect code has changed over the
years.
Reported-by: NEli Dorfman <eli@vastdata.com>
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Ensure that the re-establishment delay does not grow exponentially
on each good reconnect. This probably should have been part of
commit 675dd90a ("xprtrdma: Modernize ops->connect").
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Micro-optimization: In rpcrdma_post_recvs, since commit e340c2d6
("xprtrdma: Reduce the doorbell rate (Receive)"), the common case is
to return without doing anything. Found with perf.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

rpcrdma_rep objects are removed from their free list by only a
single thread: the Receive completion handler. Thus that free list
can be converted to an llist, where a single-threaded consumer and
a multi-threaded producer (rpcrdma_buffer_put) can both access the
llist without the need for any serialization.

This eliminates spin lock contention between the Receive completion
handler and rpcrdma_buffer_get, and makes the rep consumer wait-
free.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: Now that the free list is used sparingly, get rid of the
separate spin lock protecting it.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>