SUNRPC: Reduce thread wake-up rate when receiving large RPC messages

SUNRPC: Reduce thread wake-up rate when receiving large RPC messages

With large NFS WRITE requests on TCP, I measured 5-10 thread wake-
ups to receive each request. This is because the socket layer
calls ->sk_data_ready() frequently, and each call triggers a
thread wake-up. Each recvmsg() seems to pull in less than 100KB.

Have the socket layer hold ->sk_data_ready() calls until the full
incoming message has arrived to reduce the wake-up rate.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>