Clean up.

Remove a conditional branch from the SGL set-up loop in frwr_map():
Instead of using either sg_set_page() or sg_set_buf(), initialize
the mr_page field properly when rpcrdma_convert_kvec() converts the
kvec to an SGL entry. frwr_map() can then invoke sg_set_page()
unconditionally.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Reviewed-by: NTom Talpey <tom@talpey.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Now that rpcrdma_ep is no longer part of rpcrdma_xprt, there are
four or five serial address dereferences needed to get to the
IB device needed for DMA unmapping.

Instead, let's use the same pattern that regbufs use: cache a
pointer to the device in the MR, and use that as the indication
that unmapping is necessary.

This also guarantees that the exact same device is used for DMA
mapping and unmapping, even if the r_xprt's ep has been replaced. I
don't think this can happen today, but future changes might break
this assumption.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: This function is now invoked only in frwr_ops.c. The move
enables deduplication of the trace_xprtrdma_mr_unmap() call site.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Tie each MR event to the requesting rpc_task to make it easier to
follow MR ownership and control flow.

MR unmapping and recycling can happen in the background, after an
MR's mr_req field is stale, so set up a separate tracepoint class
for those events.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

- Replace the display of kernel memory addresses
- Add "_err" to the end of its name to indicate that it's a
  tracepoint that fires only when there's an error
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Set up a completion ID in each rpcrdma_frwr. The ID is used to match
an incoming completion to a transport (CQ) and other MR-related
activity.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

sg_init_table zeroes its first argument, so the allocation of that argument
doesn't have to.

the semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
expression x,n,flags;
@@

x =
- kcalloc
+ kmalloc_array
  (n,sizeof(*x),flags)
...
sg_init_table(x,n)
// </smpl>
Signed-off-by: NJulia Lawall <Julia.Lawall@inria.fr>
Acked-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: De-duplicate some code.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>

Refactor: Pass struct rpcrdma_xprt instead of an IB layer object.
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>

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: 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>

Two changes:
- Show the number of SG entries that were mapped. This helps debug
  DMA-related problems.
- Record the MR's resource ID instead of its memory address. This
  groups each MR with its associated rdma-tool output, and reduces
  needless exposure of memory addresses.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The @nents value that was passed to ib_dma_map_sg() has to be passed
to the matching ib_dma_unmap_sg() call. If ib_dma_map_sg() choses to
concatenate sg entries, it will return a different nents value than
it was passed.

The bug was exposed by recent changes to the AMD IOMMU driver, which
enabled sg entry concatenation.

Looking all the way back to commit 4143f34e ("xprtrdma: Port to
new memory registration API") and reviewing other kernel ULPs, it's
not clear that the frwr_map() logic was ever correct for this case.
Reported-by: NAndre Tomt <andre@tomt.net>
Suggested-by: NRobin Murphy <robin.murphy@arm.com>
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Cc: stable@vger.kernel.org
Reviewed-by: NJason Gunthorpe <jgg@mellanox.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. 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>

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>

ia->ri_id is replaced during a reconnect. The connect_worker runs
with the transport send lock held to prevent ri_id from being
dereferenced by the send_request path during this process.

Currently, however, there is no guarantee that ia->ri_id is stable
in the MR recycling worker, which operates in the background and is
not serialized with the connect_worker in any way.

But now that Local_Inv completions are being done in process
context, we can handle the recycling operation there instead of
deferring the recycling work to another process. Because the
disconnect path drains all work before allowing tear down to
proceed, it is guaranteed that Local Invalidations complete only
while the ri_id pointer is stable.
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>

When adding frwr_unmap_async way back when, I re-used the existing
trace_xprtrdma_post_send() trace point to record the return code
of ib_post_send.

Unfortunately there are some cases where re-using that trace point
causes a crash. Instead, construct a trace point specific to posting
Local Invalidate WRs that will always be safe to use in that context,
and will act as a trace log eye-catcher for Local Invalidation.

Fixes: 84756894 ("xprtrdma: Remove fr_state")
Fixes: d8099fed ("xprtrdma: Reduce context switching due ... ")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Tested-by: NBill Baker <bill.baker@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

The optimization done in "xprtrdma: Simplify rpcrdma_mr_pop" was a
bit too optimistic. MRs left over after a reconnect still need to
be recycled, not added back to the free list, since they could be
in flight or actually fully registered.
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>

Instead of a globally-contended MR free list, cache MRs in each
rpcrdma_req as they are released. This means acquiring and releasing
an MR will be lock-free in the common case, even outside the
transport send lock.

The original idea of per-rpcrdma_req MR free lists was suggested by
Shirley Ma <shirley.ma@oracle.com> several years ago. I just now
figured out how to make that idea work with on-demand MR allocation.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Probably would be good to also pass GFP flags to ib_alloc_mr.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Refactor: Retrieve an MR and handle error recovery entirely in
rpc_rdma.c, as this is not a device-specific function.

Note that since commit 89f90fe1 ("SUNRPC: Allow calls to
xprt_transmit() to drain the entire transmit queue"), the
xprt_transmit function handles the cond_resched. The transport no
longer has to do this itself.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up. There is only one remaining rpcrdma_mr_put call site, and
it can be directly replaced with unmap_and_put because mr->mr_dir is
set to DMA_NONE just before the call.

Now all the call sites do a DMA unmap, and we can just rename
mr_unmap_and_put to mr_put, which nicely matches mr_get.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: rpcrdma_mr_pop call sites check if the list is empty
first. Let's replace the list_empty with less costly logic.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Commit 302d3deb ("xprtrdma: Prevent inline overflow") added this
calculation back in 2016, but got it wrong. I tested only the lower
bound, which is why there is a max_t there. The upper bound should be
rounded up too.

Now, when using DIV_ROUND_UP, that takes care of the lower bound as
well.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Things have changed since this comment was written. In particular,
the reworking of connection closing, on-demand creation of MRs, and
the removal of fr_state all mean that deferring MR recovery to
frwr_map is no longer needed. The description is obsolete.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Since commit ba69cd12 ("xprtrdma: Remove support for FMR memory
registration"), FRWR is the only supported memory registration mode.

We can take advantage of the asynchronous nature of FRWR's LOCAL_INV
Work Requests to get rid of the completion wait by having the
LOCAL_INV completion handler take care of DMA unmapping MRs and
waking the upper layer RPC waiter.

This eliminates two context switches when local invalidation is
necessary. As a side benefit, we will no longer need the per-xprt
deferred completion work queue.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

When a marshal operation fails, any MRs that were already set up for
that request are recycled. Recycling releases MRs and creates new
ones, which is expensive.

Since commit f2877623 ("xprtrdma: Chain Send to FastReg WRs")
was merged, recycling FRWRs is unnecessary. This is because before
that commit, frwr_map had already posted FAST_REG Work Requests,
so ownership of the MRs had already been passed to the NIC and thus
dealing with them had to be delayed until they completed.

Since that commit, however, FAST_REG WRs are posted at the same time
as the Send WR. This means that if marshaling fails, we are certain
the MRs are safe to simply unmap and place back on the free list
because neither the Send nor the FAST_REG WRs have been posted yet.
The kernel still has ownership of the MRs at this point.

This reduces the total number of MRs that the xprt has to create
under heavy workloads and makes the marshaling logic less brittle.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Now that both the Send and Receive completions are handled in
process context, it is safe to DMA unmap and return MRs to the
free or recycle lists directly in the completion handlers.

Doing this means rpcrdma_frwr no longer needs to track the state of
each MR, meaning that a VALID or FLUSHED MR can no longer appear on
an xprt's MR free list. Thus there is no longer a need to track the
MR's registration state in rpcrdma_frwr.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Under high I/O workloads, I've noticed that an RPC/RDMA transport
occasionally deadlocks (IOPS goes to zero, and doesn't recover).
Diagnosis shows that the sendctx queue is empty, but when sendctxs
are returned to the queue, the xprt_write_space wake-up never
occurs. The wake-up logic in rpcrdma_sendctx_put_locked is racy.

I noticed that both EMPTY_SCQ and XPRT_WRITE_SPACE are implemented
via an atomic bit. Just one of those is sufficient. Removing
EMPTY_SCQ in favor of the generic bit mechanism makes the deadlock
un-reproducible.

Without EMPTY_SCQ, rpcrdma_buffer::rb_flags is no longer used and
is therefore removed.

Unfortunately this patch does not apply cleanly to stable. If
needed, someone will have to port it and test it.

Fixes: 2fad6592 ("xprtrdma: Wait on empty sendctx queue")
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up: rely on the trace points instead.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up.

Move the remaining field in rpcrdma_create_data_internal so the
structure can be removed.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

Clean up.

Since commit 54cbd6b0 ("xprtrdma: Delay DMA mapping Send and
Receive buffers"), a pointer to the device is now saved in each
regbuf when it is DMA mapped.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>

After a DMA map failure in frwr_map, mark the MR so that recycling
won't attempt to DMA unmap it.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Fixes: e2f34e26 ("xprtrdma: Yet another double DMA-unmap")
Signed-off-by: NAnna Schumaker <Anna.Schumaker@Netapp.com>