OSDN Git Service

xprtrdma: Move Receive posting to Receive handler
authorChuck Lever <chuck.lever@oracle.com>
Fri, 4 May 2018 19:35:20 +0000 (15:35 -0400)
committerAnna Schumaker <Anna.Schumaker@Netapp.com>
Mon, 7 May 2018 13:20:04 +0000 (09:20 -0400)
commit7c8d9e7c8863905951d4eaa7a8d277150f3a37f7
tree16017a18ac2150272b7b5dd4be6fe1cd5d230aeb
parent0e0b854cfb3302b1907e9d3a927469b95710238f
xprtrdma: Move Receive posting to Receive handler

Receive completion and Reply handling are done by a BOUND
workqueue, meaning they run on only one CPU.

Posting receives is currently done in the send_request path, which
on large systems is typically done on a different CPU than the one
handling Receive completions. This results in movement of
Receive-related cachelines between the sending and receiving CPUs.

More importantly, it means that currently Receives are posted while
the transport's write lock is held, which is unnecessary and costly.

Finally, allocation of Receive buffers is performed on-demand in
the Receive completion handler. This helps guarantee that they are
allocated on the same NUMA node as the CPU that handles Receive
completions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
include/trace/events/rpcrdma.h
net/sunrpc/xprtrdma/backchannel.c
net/sunrpc/xprtrdma/rpc_rdma.c
net/sunrpc/xprtrdma/transport.c
net/sunrpc/xprtrdma/verbs.c
net/sunrpc/xprtrdma/xprt_rdma.h