[dpdk-dev] [PATCH] event/sw: fix hashing of flow on ordered ingress

Harry van Haaren harry.van.haaren at intel.com
Mon Apr 10 17:56:43 CEST 2017


The flow id of packets was not being hashed on ingress
on an ordered queue. Fix by applying same hashing as is
applied in the atomic queue case. The hashing itself is
broken out into a macro to avoid duplication of code.

Fixes: 617995dfc5b2 ("event/sw: add scheduling logic")

Signed-off-by: Harry van Haaren <harry.van.haaren at intel.com>
---
 drivers/event/sw/sw_evdev_scheduler.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/drivers/event/sw/sw_evdev_scheduler.c b/drivers/event/sw/sw_evdev_scheduler.c
index 77a16d7..e008b51 100644
--- a/drivers/event/sw/sw_evdev_scheduler.c
+++ b/drivers/event/sw/sw_evdev_scheduler.c
@@ -51,6 +51,8 @@
 
 #define MAX_PER_IQ_DEQUEUE 48
 #define FLOWID_MASK (SW_QID_NUM_FIDS-1)
+/* use cheap bit mixing, we only need to lose a few bits */
+#define SW_HASH_FLOWID(f) (((f) ^ (f >> 10)) & FLOWID_MASK)
 
 static inline uint32_t
 sw_schedule_atomic_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
@@ -72,9 +74,7 @@ sw_schedule_atomic_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
 	iq_ring_dequeue_burst(qid->iq[iq_num], qes, count);
 	for (i = 0; i < count; i++) {
 		const struct rte_event *qe = &qes[i];
-		/* use cheap bit mixing, we only need to lose a few bits */
-		uint32_t flow_id32 = (qes[i].flow_id) ^ (qes[i].flow_id >> 10);
-		const uint16_t flow_id = FLOWID_MASK & flow_id32;
+		const uint16_t flow_id = SW_HASH_FLOWID(qes[i].flow_id);
 		struct sw_fid_t *fid = &qid->fids[flow_id];
 		int cq = fid->cq;
 
@@ -183,8 +183,7 @@ sw_schedule_parallel_to_cq(struct sw_evdev *sw, struct sw_qid * const qid,
 		qid->stats.tx_pkts++;
 
 		const int head = (p->hist_head & (SW_PORT_HIST_LIST-1));
-
-		p->hist_list[head].fid = qe->flow_id;
+		p->hist_list[head].fid = SW_HASH_FLOWID(qe->flow_id);
 		p->hist_list[head].qid = qid_id;
 
 		if (keep_order)
-- 
2.7.4



More information about the dev mailing list