From: NeilBrown <neilb@suse.de>
Subject: NFS: mark nfsiod as CPU_INTENSIVE
References: bsc#1177304
Patch-mainline: never, maintainer wants are more general solution

Very occasionally nfsiod will be called on to call iput_final() on an
inode.  If that inode has millions of pages it can take a while to fully
clean up.  This blocks all other wq workers on that CPU, despite the
fact that cond_resync() is being called.

If we mark the workqueue CPU_INTENSIVE, those workers can run in
parallel with the iput_final().

This is likely to increase the number of worker threads but as most work
items complete extremely quickly, it won't increase it much.

Signed-off-by: NeilBrown <neilb@suse.de>
---
 fs/nfs/inode.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -2025,7 +2025,7 @@ static int nfsiod_start(void)
 {
 	struct workqueue_struct *wq;
 	dprintk("RPC:       creating workqueue nfsiod\n");
-	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
+	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_CPU_INTENSIVE, 0);
 	if (wq == NULL)
 		return -ENOMEM;
 	nfsiod_workqueue = wq;