Performance regression reported with NFS/RDMA using Omnipath,
bisected to commit e084ee673c ("svcrdma: Add Write chunk WRs to
the RPC's Send WR chain").
Tracing on the server reports:
nfsd-7771 [060] 1758.891809: svcrdma_sq_post_err:
cq.id=205 cid=226 sc_sq_avail=13643/851 status=-12
sq_post_err reports ENOMEM, and the rdma->sc_sq_avail (13643) is
larger than rdma->sc_sq_depth (851). The number of available Send
Queue entries is always supposed to be smaller than the Send Queue
depth. That seems like a Send Queue accounting bug in svcrdma.
As it's getting to be late in the 6.9-rc cycle, revert this commit.
It can be revisited in a subsequent kernel release.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=218743
Fixes: e084ee673c ("svcrdma: Add Write chunk WRs to the RPC's Send WR chain")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Chain RDMA Writes that convey Write chunks onto the local Send
chain. This means all WRs for an RPC Reply are now posted with a
single ib_post_send() call, and there is a single Send completion
when all of these are done. That reduces both the per-transport
doorbell rate and completion rate.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Refactor to eventually enable svcrdma to post the Write WRs for each
RPC response using the same ib_post_send() as the Send WR (ie, as a
single WR chain).
svc_rdma_result_payload (originally svc_rdma_read_payload) was added
so that the upper layer XDR encoder could identify a range of bytes
to be possibly conveyed by RDMA (if a Write chunk was provided by
the client).
The purpose of commit f6ad77590a ("svcrdma: Post RDMA Writes while
XDR encoding replies") was to post as much of the result payload
outside of svc_rdma_sendto() as possible because svc_rdma_sendto()
used to be called with the xpt_mutex held.
However, since commit ca4faf543a ("SUNRPC: Move xpt_mutex into
socket xpo_sendto methods"), the xpt_mutex is no longer held when
calling svc_rdma_sendto(). Thus, that benefit is no longer an issue.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reduce the doorbell and Send completion rates when sending RPC/RDMA
replies that have Reply chunks. NFS READDIR procedures typically
return their result in a Reply chunk, for example.
Instead of calling ib_post_send() to post the Write WRs for the
Reply chunk, and then calling it again to post the Send WR that
conveys the transport header, chain the Write WRs to the Send WR
and call ib_post_send() only once.
Thanks to the Send Queue completion ordering rules, when the Send
WR completes, that guarantees that Write WRs posted before it have
also completed successfully. Thus all Write WRs for the Reply chunk
can remain unsignaled. Instead of handling a Write completion and
then a Send completion, only the Send completion is seen, and it
handles clean up for both the Writes and the Send.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Since the RPC transaction's svc_rdma_send_ctxt will stay around for
the duration of the RDMA Write operation, the write_info structure
for the Reply chunk can reside in the request's svc_rdma_send_ctxt
instead of being allocated separately.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Eventually I'd like the server to post the reply's Send WR along
with any Write WRs using only a single call to ib_post_send(), in
order to reduce the NIC's doorbell rate.
To do this, add an anchor for a WR chain to svc_rdma_send_ctxt, and
refactor svc_rdma_send() to post this WR chain to the Send Queue. For
the moment, the posted chain will continue to contain a single Send
WR.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
In some error flow cases, svc_rdma_wc_send() releases @ctxt. Copy
the sc_cid field in @ctxt to a stack variable in order to guarantee
that the value is available after the ib_post_send() call.
In case the new comment looks a little strange, this will be done
with at least one more field in a subsequent patch.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Ensure there is a wake-up when increasing sc_sq_avail.
Likewise, if a wake-up is done, sc_sq_avail needs to be updated,
otherwise the wait_event() conditional is never going to be met.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The atomic_inc_return() in svc_rdma_send_cid_init() is expensive.
Some svc_rdma_chunk_ctxt's now reside in long-lived container
structures. They don't need a fresh completion ID for every I/O
operation.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
A send/recv_ctxt already records transport-related information
in the cq.id, thus there is no need to record the IP addresses of
the transport endpoints.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Update the Send Queue's error flow tracepoints to report the
completion ID of the waiting or failing context. This ties the
wait/failure to a particular operation or request, which is a little
more useful than knowing only the transport that is about to close.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
DMA unmapping can take quite some time, so it should not be handled
in a single-threaded completion handler. Defer releasing send_ctxts
to the recently-added workqueue.
With this patch, DMA unmapping can be handled in parallel, and it
does not cause head-of-queue blocking of Send completions.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The svc_rdma_recv_ctxt free list uses a lockless list to avoid the
need for a spin lock in the fast path. llist_del_first(), which is
used by svc_rdma_recv_ctxt_get(), requires serialization, however,
when there are multiple list producers that are unserialized.
I mistakenly thought there was only one caller of
svc_rdma_recv_ctxt_get() (svc_rdma_refresh_recvs()), thus explicit
serialization would not be necessary. But there is another caller:
svc_rdma_bc_sendto(), and these two are not serialized against each
other. I haven't seen ill effects that I could directly ascribe to
a lack of serialization. It's just an observation based on code
audit.
When DMA-mapping before sending a Reply, the passed-in struct
svc_rdma_recv_ctxt is used only for its write and reply PCLs. These
are currently always empty in the backchannel case. So, instead of
passing a full svc_rdma_recv_ctxt object to
svc_rdma_map_reply_msg(), let's pass in just the Write and Reply
PCLs.
This change makes it unnecessary for the backchannel to acquire a
dummy svc_rdma_recv_ctxt object when sending an RPC Call. The need
for svc_rdma_recv_ctxt free list serialization is now completely
avoided.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Try to catch incorrect calling contexts mechanically rather than by
code review.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Now that we have bulk page allocation and release APIs, it's more
efficient to use those than it is for nfsd threads to wait for send
completions. Previous patches have eliminated the calls to
wait_for_completion() and complete(), in order to avoid scheduler
overhead.
Now release pages-under-I/O in the send completion handler using
the efficient bulk release API.
I've measured a 7% reduction in cumulative CPU utilization in
svc_rdma_sendto(), svc_rdma_wc_send(), and svc_xprt_release(). In
particular, using release_pages() instead of complete() cuts the
time per svc_rdma_wc_send() call by two-thirds. This helps improve
scalability because svc_rdma_wc_send() is single-threaded per
connection.
Reviewed-by: Tom Talpey <tom@talpey.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Get rid of the completion wait in svc_rdma_sendto(), and release
pages in the send completion handler again. A subsequent patch will
handle releasing those pages more efficiently.
Reverted by hand: patch -R would not apply cleanly.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Pre-requisite for releasing pages in the send completion handler.
Reverted by hand: patch -R would not apply cleanly.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Pre-requisite for releasing pages in the send completion handler.
Reverted by hand: patch -R would not apply cleanly.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The physical device's favored NUMA node ID is available when
allocating a send_ctxt. Use that value instead of relying on the
assumption that the memory allocation happens to be running on a
node close to the device.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There are currently three separate purposes being served by a single
tracepoint here. They need to be split up.
svcrdma_wc_send:
- status is always zero, so there's no value in recording it.
- vendor_err is meaningless unless status is not zero, so
there's no value in recording it.
- This tracepoint is needed only when developing modifications,
so it should be left disabled most of the time.
svcrdma_wc_send_flush:
- As above, needed only rarely, and not an error.
svcrdma_wc_send_err:
- This tracepoint can be left persistently enabled because
completion errors are run-time problems (except for FLUSHED_ERR).
- Tracepoint name now ends in _err to reflect its purpose.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
/proc/lock_stat indicates the the sc_send_lock is heavily
contended when the server is under load from a single client.
To address this, convert the send_ctxt free list to an llist.
Returning an item to the send_ctxt cache is now waitless, which
reduces the instruction path length in the single-threaded Send
handler (svc_rdma_wc_send).
The goal is to enable the ib_comp_wq worker to handle a higher
RPC/RDMA Send completion rate given the same CPU resources. This
change reduces CPU utilization of Send completion by 2-3% on my
server.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-By: Tom Talpey <tom@talpey.com>
Because wake_up() takes an IRQ-safe lock, it can be expensive,
especially to call inside of a single-threaded completion handler.
What's more, the Send wait queue almost never has waiters, so
most of the time, this is an expensive no-op.
As always, the goal is to reduce the average overhead of each
completion, because a transport's completion handlers are single-
threaded on one CPU core. This change reduces CPU utilization of
the Send completion thread by 2-3% on my server.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-By: Tom Talpey <tom@talpey.com>
Capture error codes in @ret, which is passed to the send_err
tracepoint, so that they can be logged when something goes awry.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
svc_rdma_sendto() now waits for the NIC hardware to finish with
the pages backing rq_res. We still have to release the page array
in some cases, but now it's always safe to immediately re-use the
page backing rq_res's head buffer.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Currently svc_rdma_sendto() migrates xdr_buf pages into a separate
page list and NULLs out a bunch of entries in rq_pages while the
pages are under I/O. The Send completion handler then frees those
pages later.
Instead, let's wait for the Send completion, then handle page
releasing in the nfsd thread. I'd like to avoid the cost of 250+
put_page() calls in the Send completion handler, which is single-
threaded.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Refactor a bit of commonly used logic so that every site that wants
a close deferred to an nfsd thread does all the right things
(set_bit(XPT_CLOSE) then enqueue).
Also, once XPT_CLOSE is set on a transport, it is never cleared. If
XPT_CLOSE is already set, then the close is already being handled
and the enqueue can be skipped.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
We already have trace_svcrdma_decode_rseg(), which records each
ingress Read segment. Instead of reporting those again when they
are about to be posted as RDMA Reads, let's fire one tracepoint
before posting each type of chunk.
So we'll get:
nfsd-1998 [002] 321.666615: svcrdma_decode_rseg: cq.id=4 cid=42 segno=0 position=0 192@0x013ca9ebfae14000:0xb0010b05
nfsd-1998 [002] 321.666615: svcrdma_decode_rseg: cq.id=4 cid=42 segno=1 position=0 7688@0x013ca9ebf914e000:0xb0010a05
nfsd-1998 [002] 321.666615: svcrdma_decode_rseg: cq.id=4 cid=42 segno=2 position=0 28@0x013ca9ebfae15000:0xb0010905
nfsd-1998 [002] 321.666622: svcrdma_decode_rqst: cq.id=4 cid=42 xid=0x013ca9eb vers=1 credits=128 proc=RDMA_NOMSG hdrlen=100
nfsd-1998 [002] 321.666642: svcrdma_post_read_chunk: cq.id=3 cid=112 sqecount=3
kworker/2:1H-221 [002] 321.673949: svcrdma_wc_read: cq.id=3 cid=112 status=SUCCESS (0/0x0)
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Refactor: svc_rdma_map_reply_msg() is restructured to DMA map only
the parts of rq_res that do not contain a result payload.
This change has been tested to confirm that it does not cause a
regression in the no Write chunk and single Write chunk cases.
Multiple Write chunks have not been tested.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When counting the number of SGEs needed to construct a Send request,
do not count result payloads. And, when copying the Reply message
into the pull-up buffer, result payloads are not to be copied to the
Send buffer.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Refactor: Instead of re-parsing the ingress RPC Call transport
header when constructing the egress RPC Reply transport header, use
the new parsed Write list and Reply chunk, which are version-
agnostic and already XDR decoded.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Refactor: Instead of re-parsing the ingress RPC Call transport
header when constructing RDMA Writes, use the new parsed chunk lists
for the Write list and Reply chunk, which are version-agnostic and
already XDR-decoded.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The only RPC/RDMA ordering requirement between RDMA Writes and RDMA
Sends is that the responder must post the Writes on the Send queue
before posting the Send that conveys the RPC Reply for that Write
payload.
The Linux NFS server implementation now has a transport method that
can post result Payload Writes earlier than svc_rdma_sendto:
->xpo_result_payload()
This gets RDMA Writes going earlier so they are more likely to be
complete at the remote end before the Send completes.
Some care must be taken with pulled-up Replies. We don't want to
push the Write chunk and then send the same payload data via Send.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Have the NFSD encoders annotate the boundaries of every
direct-data-placement eligible result data payload. Then change
svcrdma to use that annotation instead of the xdr->page_len
when handling Write chunks.
For NFSv4 on RDMA, that enables the ability to recognize multiple
result payloads per compound. This is a pre-requisite for supporting
multiple Write chunks per RPC transaction.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Clean up: "result payload" is a less confusing name for these
payloads. "READ payload" reflects only the NFS usage.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This was discovered using O_DIRECT at the client side, with small
unaligned file offsets or IOs that span multiple file pages.
Fixes: e248aa7be8 ("svcrdma: Remove max_sge check at connect time")
Signed-off-by: Dan Aloni <dan@kernelim.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Jason tells me that a ULP cannot rely on getting an ESTABLISHED
and DISCONNECTED event pair for each connection, so transport
reference counting in the CM event handler will never be reliable.
Now that we have ib_drain_qp(), svcrdma should no longer need to
hold transport references while Sends and Receives are posted. So
remove the get/put call sites in the CM event handlers.
This eliminates a significant source of locked memory bus traffic.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
First, refactor: Dereference the svc_rdma_send_ctxt inside
svc_rdma_send() instead of at every call site.
Then, it can be passed into trace_svcrdma_post_send() to get the
proper completion ID.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Set up a completion ID in each svc_rdma_send_ctxt. The ID is used
to match an incoming Send completion to a transport and to a
previous ib_post_send().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Chuck Lever