staging/lustre/ptlrpc: avoid list scan in ptlrpcd_check

ptlrpcd_check() always scan all requests on ptlrpc_request_set
and try to finish completed requests, this is low efficiency.
Even worse, l_wait_event() always checks condition for twice
before sleeping and one more time after waking up, which means
it will call ptlrpcd_check() for three times in each loop.

This patch will move completed requests at the head of list
in ptlrpc_check_set(), with this change ptlrpcd_check doesn't
need to scan all requests anymore.

Signed-off-by: Liang Zhen <liang.zhen@intel.com>
Reviewed-on: http://review.whamcloud.com/11513
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5548
Reviewed-by: Bobi Jam <bobijam@gmail.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Johann Lombardi <johann.lombardi@intel.com>
Signed-off-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

2 files changed, 20 insertions, 15 deletions

diff --git a/drivers/staging/lustre/lustre/ptlrpc/client.c b/drivers/staging/lustre/lustre/ptlrpc/client.c
index dc9e406f3212..8c1ec830712f 100644
--- a/drivers/staging/lustre/lustre/ptlrpc/client.c
+++ b/drivers/staging/lustre/lustre/ptlrpc/client.c
@@ -1497,11 +1497,13 @@ static inline int ptlrpc_set_producer(struct ptlrpc_request_set *set)
 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set)
 {
 	struct list_head *tmp, *next;
+	struct list_head comp_reqs;
 	int force_timer_recalc = 0;
 
 	if (atomic_read(&set->set_remaining) == 0)
 		return 1;
 
+	INIT_LIST_HEAD(&comp_reqs);
 	list_for_each_safe(tmp, next, &set->set_requests) {
 		struct ptlrpc_request *req =
 			list_entry(tmp, struct ptlrpc_request,
@@ -1576,8 +1578,10 @@ int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set)
 			ptlrpc_rqphase_move(req, req->rq_next_phase);
 		}
 
-		if (req->rq_phase == RQ_PHASE_COMPLETE)
+		if (req->rq_phase == RQ_PHASE_COMPLETE) {
+			list_move_tail(&req->rq_set_chain, &comp_reqs);
 			continue;
+		}
 
 		if (req->rq_phase == RQ_PHASE_INTERPRET)
 			goto interpret;
@@ -1860,9 +1864,15 @@ interpret:
 			if (req->rq_status != 0)
 				set->set_rc = req->rq_status;
 			ptlrpc_req_finished(req);
+		} else {
+			list_move_tail(&req->rq_set_chain, &comp_reqs);
 		}
 	}
 
+	/* move completed request at the head of list so it's easier for
+	 * caller to find them */
+	list_splice(&comp_reqs, &set->set_requests);
+
 	/* If we hit an error, we want to recover promptly. */
 	return atomic_read(&set->set_remaining) == 0 || force_timer_recalc;
 }
diff --git a/drivers/staging/lustre/lustre/ptlrpc/ptlrpcd.c b/drivers/staging/lustre/lustre/ptlrpc/ptlrpcd.c
index cbcc541cac43..4621b71fe0b6 100644
--- a/drivers/staging/lustre/lustre/ptlrpc/ptlrpcd.c
+++ b/drivers/staging/lustre/lustre/ptlrpc/ptlrpcd.c
@@ -306,21 +306,16 @@ static int ptlrpcd_check(struct lu_env *env, struct ptlrpcd_ctl *pc)
 	if (atomic_read(&set->set_remaining))
 		rc |= ptlrpc_check_set(env, set);
 
-	if (!list_empty(&set->set_requests)) {
-		/*
-		 * XXX: our set never completes, so we prune the completed
-		 * reqs after each iteration. boy could this be smarter.
-		 */
-		list_for_each_safe(pos, tmp, &set->set_requests) {
-			req = list_entry(pos, struct ptlrpc_request,
-					     rq_set_chain);
-			if (req->rq_phase != RQ_PHASE_COMPLETE)
-				continue;
+	/* NB: ptlrpc_check_set has already moved completed request at the
+	 * head of seq::set_requests */
+	list_for_each_safe(pos, tmp, &set->set_requests) {
+		req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
+		if (req->rq_phase != RQ_PHASE_COMPLETE)
+			break;
 
-			list_del_init(&req->rq_set_chain);
-			req->rq_set = NULL;
-			ptlrpc_req_finished(req);
-		}
+		list_del_init(&req->rq_set_chain);
+		req->rq_set = NULL;
+		ptlrpc_req_finished(req);
 	}
 
 	if (rc == 0) {