fix parallelism for rpc tasks

Hi folks,

On a multi-core machine, is it expected that we can have parallel RPCs
handled by each of the per-core workqueue?

In testing a read workload, observing via "top" command that a single
"kworker" thread is running servicing the requests (no parallelism).
It's more prominent while doing these operations over krb5p mount.

What has been suggested by Bruce is to try this and in my testing I
see then the read workload spread among all the kworker threads.

Signed-off-by: Olga Kornievskaia <[email protected]>
Signed-off-by: Trond Myklebust <[email protected]>

Author: olgakorn

net/sunrpc/sched.c

@@ -1104,7 +1104,7 @@ static int rpciod_start(void)
 	 * Create the rpciod thread and wait for it to start.
 	 */
 	dprintk("RPC:       creating workqueue rpciod\n");
-	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
+	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 	if (!wq)
 		goto out_failed;
 	rpciod_workqueue = wq;