[dpdk-dev] [PATCH v7 03/10] examples/ipsec-secgw: fix crypto-op might never get dequeued

[Konstantin Ananyev <konstantin.ananyev@intel.com>]

In some cases crypto-ops could never be dequeued from the crypto-device.
The easiest way to reproduce:
start ipsec-secgw with crypto-dev and send to it less then 32 packets.
none packets will be forwarded.
Reason for that is that the application does dequeue() from crypto-queues
only when new packets arrive.
This patch makes the app to call dequeue() on a regular basis.
Also to make code cleaner and easier to understand,
it separates crypto-dev enqueue() and dequeue() code paths.
pkt_process() now only enqueues packets into crypto device,
dequeuing and final processing is done by drain_crypto_queues().

Fixes: c64278c0c18b ("examples/ipsec-secgw: rework processing loop")
Cc: stable@dpdk.org

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Radu Nicolau <radu.nicolau@intel.com>
---
 examples/ipsec-secgw/ipsec-secgw.c | 144 ++++++++++++++++++++++++-----
 examples/ipsec-secgw/ipsec.c       |  99 +++++++++++++-------
 examples/ipsec-secgw/ipsec.h       |  11 +++
 3 files changed, 198 insertions(+), 56 deletions(-)

diff --git a/examples/ipsec-secgw/ipsec-secgw.c b/examples/ipsec-secgw/ipsec-secgw.c
index e79d17d51..f8290c13c 100644
--- a/examples/ipsec-secgw/ipsec-secgw.c
+++ b/examples/ipsec-secgw/ipsec-secgw.c
@@ -469,38 +469,55 @@ inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 	ip->num = j;
 }
 
-static inline void
-process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
-		struct ipsec_traffic *traffic)
+static void
+split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 {
+	uint32_t i, n4, n6;
+	struct ip *ip;
 	struct rte_mbuf *m;
-	uint16_t idx, nb_pkts_in, i, n_ip4, n_ip6;
 
-	nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
-			traffic->ipsec.num, MAX_PKT_BURST);
+	n4 = trf->ip4.num;
+	n6 = trf->ip6.num;
 
-	n_ip4 = traffic->ip4.num;
-	n_ip6 = traffic->ip6.num;
+	for (i = 0; i < num; i++) {
+
+		m = mb[i];
+		ip = rte_pktmbuf_mtod(m, struct ip *);
 
-	/* SP/ACL Inbound check ipsec and ip4 */
-	for (i = 0; i < nb_pkts_in; i++) {
-		m = traffic->ipsec.pkts[i];
-		struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 		if (ip->ip_v == IPVERSION) {
-			idx = traffic->ip4.num++;
-			traffic->ip4.pkts[idx] = m;
-			traffic->ip4.data[idx] = rte_pktmbuf_mtod_offset(m,
+			trf->ip4.pkts[n4] = m;
+			trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 					uint8_t *, offsetof(struct ip, ip_p));
+			n4++;
 		} else if (ip->ip_v == IP6_VERSION) {
-			idx = traffic->ip6.num++;
-			traffic->ip6.pkts[idx] = m;
-			traffic->ip6.data[idx] = rte_pktmbuf_mtod_offset(m,
+			trf->ip6.pkts[n6] = m;
+			trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 					uint8_t *,
 					offsetof(struct ip6_hdr, ip6_nxt));
+			n6++;
 		} else
 			rte_pktmbuf_free(m);
 	}
 
+	trf->ip4.num = n4;
+	trf->ip6.num = n6;
+}
+
+
+static inline void
+process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
+		struct ipsec_traffic *traffic)
+{
+	uint16_t nb_pkts_in, n_ip4, n_ip6;
+
+	n_ip4 = traffic->ip4.num;
+	n_ip6 = traffic->ip6.num;
+
+	nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
+			traffic->ipsec.num, MAX_PKT_BURST);
+
+	split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
+
 	inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 			n_ip4);
 
@@ -795,7 +812,7 @@ process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 }
 
 static inline void
-drain_buffers(struct lcore_conf *qconf)
+drain_tx_buffers(struct lcore_conf *qconf)
 {
 	struct buffer *buf;
 	uint32_t portid;
@@ -809,6 +826,81 @@ drain_buffers(struct lcore_conf *qconf)
 	}
 }
 
+static inline void
+drain_crypto_buffers(struct lcore_conf *qconf)
+{
+	uint32_t i;
+	struct ipsec_ctx *ctx;
+
+	/* drain inbound buffers*/
+	ctx = &qconf->inbound;
+	for (i = 0; i != ctx->nb_qps; i++) {
+		if (ctx->tbl[i].len != 0)
+			enqueue_cop_burst(ctx->tbl  + i);
+	}
+
+	/* drain outbound buffers*/
+	ctx = &qconf->outbound;
+	for (i = 0; i != ctx->nb_qps; i++) {
+		if (ctx->tbl[i].len != 0)
+			enqueue_cop_burst(ctx->tbl  + i);
+	}
+}
+
+static void
+drain_inbound_crypto_queues(const struct lcore_conf *qconf,
+		struct ipsec_ctx *ctx)
+{
+	uint32_t n;
+	struct ipsec_traffic trf;
+
+	/* dequeue packets from crypto-queue */
+	n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
+			RTE_DIM(trf.ipsec.pkts));
+	if (n == 0)
+		return;
+
+	trf.ip4.num = 0;
+	trf.ip6.num = 0;
+
+	/* split traffic by ipv4-ipv6 */
+	split46_traffic(&trf, trf.ipsec.pkts, n);
+
+	/* process ipv4 packets */
+	inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
+	route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
+
+	/* process ipv6 packets */
+	inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
+	route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
+}
+
+static void
+drain_outbound_crypto_queues(const struct lcore_conf *qconf,
+		struct ipsec_ctx *ctx)
+{
+	uint32_t n;
+	struct ipsec_traffic trf;
+
+	/* dequeue packets from crypto-queue */
+	n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
+			RTE_DIM(trf.ipsec.pkts));
+	if (n == 0)
+		return;
+
+	trf.ip4.num = 0;
+	trf.ip6.num = 0;
+
+	/* split traffic by ipv4-ipv6 */
+	split46_traffic(&trf, trf.ipsec.pkts, n);
+
+	/* process ipv4 packets */
+	route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
+
+	/* process ipv6 packets */
+	route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
+}
+
 /* main processing loop */
 static int32_t
 main_loop(__attribute__((unused)) void *dummy)
@@ -866,12 +958,14 @@ main_loop(__attribute__((unused)) void *dummy)
 		diff_tsc = cur_tsc - prev_tsc;
 
 		if (unlikely(diff_tsc > drain_tsc)) {
-			drain_buffers(qconf);
+			drain_tx_buffers(qconf);
+			drain_crypto_buffers(qconf);
 			prev_tsc = cur_tsc;
 		}
 
-		/* Read packet from RX queues */
 		for (i = 0; i < qconf->nb_rx_queue; ++i) {
+
+			/* Read packets from RX queues */
 			portid = rxql[i].port_id;
 			queueid = rxql[i].queue_id;
 			nb_rx = rte_eth_rx_burst(portid, queueid,
@@ -879,6 +973,14 @@ main_loop(__attribute__((unused)) void *dummy)
 
 			if (nb_rx > 0)
 				process_pkts(qconf, pkts, nb_rx, portid);
+
+			/* dequeue and process completed crypto-ops */
+			if (UNPROTECTED_PORT(portid))
+				drain_inbound_crypto_queues(qconf,
+					&qconf->inbound);
+			else
+				drain_outbound_crypto_queues(qconf,
+					&qconf->outbound);
 		}
 	}
 }
diff --git a/examples/ipsec-secgw/ipsec.c b/examples/ipsec-secgw/ipsec.c
index 3d415f1af..72a29bcb1 100644
--- a/examples/ipsec-secgw/ipsec.c
+++ b/examples/ipsec-secgw/ipsec.c
@@ -333,33 +333,35 @@ create_session(struct ipsec_ctx *ipsec_ctx, struct ipsec_sa *sa)
 	return 0;
 }
 
+/*
+ * queue crypto-ops into PMD queue.
+ */
+void
+enqueue_cop_burst(struct cdev_qp *cqp)
+{
+	uint32_t i, len, ret;
+
+	len = cqp->len;
+	ret = rte_cryptodev_enqueue_burst(cqp->id, cqp->qp, cqp->buf, len);
+	if (ret < len) {
+		RTE_LOG_DP(DEBUG, IPSEC, "Cryptodev %u queue %u:"
+			" enqueued %u crypto ops out of %u\n",
+			cqp->id, cqp->qp, ret, len);
+			/* drop packets that we fail to enqueue */
+			for (i = ret; i < len; i++)
+				rte_pktmbuf_free(cqp->buf[i]->sym->m_src);
+	}
+	cqp->in_flight += ret;
+	cqp->len = 0;
+}
+
 static inline void
 enqueue_cop(struct cdev_qp *cqp, struct rte_crypto_op *cop)
 {
-	int32_t ret = 0, i;
-
 	cqp->buf[cqp->len++] = cop;
 
-	if (cqp->len == MAX_PKT_BURST) {
-		int enq_size = cqp->len;
-		if ((cqp->in_flight + enq_size) > MAX_INFLIGHT)
-			enq_size -=
-			    (int)((cqp->in_flight + enq_size) - MAX_INFLIGHT);
-
-		if (enq_size > 0)
-			ret = rte_cryptodev_enqueue_burst(cqp->id, cqp->qp,
-					cqp->buf, enq_size);
-		if (ret < cqp->len) {
-			RTE_LOG_DP(DEBUG, IPSEC, "Cryptodev %u queue %u:"
-					" enqueued %u crypto ops out of %u\n",
-					 cqp->id, cqp->qp,
-					 ret, cqp->len);
-			for (i = ret; i < cqp->len; i++)
-				rte_pktmbuf_free(cqp->buf[i]->sym->m_src);
-		}
-		cqp->in_flight += ret;
-		cqp->len = 0;
-	}
+	if (cqp->len == MAX_PKT_BURST)
+		enqueue_cop_burst(cqp);
 }
 
 static inline void
@@ -473,6 +475,32 @@ ipsec_enqueue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
 	}
 }
 
+static inline int32_t
+ipsec_inline_dequeue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
+	      struct rte_mbuf *pkts[], uint16_t max_pkts)
+{
+	int32_t nb_pkts, ret;
+	struct ipsec_mbuf_metadata *priv;
+	struct ipsec_sa *sa;
+	struct rte_mbuf *pkt;
+
+	nb_pkts = 0;
+	while (ipsec_ctx->ol_pkts_cnt > 0 && nb_pkts < max_pkts) {
+		pkt = ipsec_ctx->ol_pkts[--ipsec_ctx->ol_pkts_cnt];
+		rte_prefetch0(pkt);
+		priv = get_priv(pkt);
+		sa = priv->sa;
+		ret = xform_func(pkt, sa, &priv->cop);
+		if (unlikely(ret)) {
+			rte_pktmbuf_free(pkt);
+			continue;
+		}
+		pkts[nb_pkts++] = pkt;
+	}
+
+	return nb_pkts;
+}
+
 static inline int
 ipsec_dequeue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
 	      struct rte_mbuf *pkts[], uint16_t max_pkts)
@@ -490,19 +518,6 @@ ipsec_dequeue(ipsec_xform_fn xform_func, struct ipsec_ctx *ipsec_ctx,
 		if (ipsec_ctx->last_qp == ipsec_ctx->nb_qps)
 			ipsec_ctx->last_qp %= ipsec_ctx->nb_qps;
 
-		while (ipsec_ctx->ol_pkts_cnt > 0 && nb_pkts < max_pkts) {
-			pkt = ipsec_ctx->ol_pkts[--ipsec_ctx->ol_pkts_cnt];
-			rte_prefetch0(pkt);
-			priv = get_priv(pkt);
-			sa = priv->sa;
-			ret = xform_func(pkt, sa, &priv->cop);
-			if (unlikely(ret)) {
-				rte_pktmbuf_free(pkt);
-				continue;
-			}
-			pkts[nb_pkts++] = pkt;
-		}
-
 		if (cqp->in_flight == 0)
 			continue;
 
@@ -545,6 +560,13 @@ ipsec_inbound(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
 
 	ipsec_enqueue(esp_inbound, ctx, pkts, sas, nb_pkts);
 
+	return ipsec_inline_dequeue(esp_inbound_post, ctx, pkts, len);
+}
+
+uint16_t
+ipsec_inbound_cqp_dequeue(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
+		uint16_t len)
+{
 	return ipsec_dequeue(esp_inbound_post, ctx, pkts, len);
 }
 
@@ -558,5 +580,12 @@ ipsec_outbound(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
 
 	ipsec_enqueue(esp_outbound, ctx, pkts, sas, nb_pkts);
 
+	return ipsec_inline_dequeue(esp_outbound_post, ctx, pkts, len);
+}
+
+uint16_t
+ipsec_outbound_cqp_dequeue(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
+		uint16_t len)
+{
 	return ipsec_dequeue(esp_outbound_post, ctx, pkts, len);
 }
diff --git a/examples/ipsec-secgw/ipsec.h b/examples/ipsec-secgw/ipsec.h
index 580f7876b..2f04b7d68 100644
--- a/examples/ipsec-secgw/ipsec.h
+++ b/examples/ipsec-secgw/ipsec.h
@@ -184,6 +184,14 @@ uint16_t
 ipsec_outbound(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
 		uint32_t sa_idx[], uint16_t nb_pkts, uint16_t len);
 
+uint16_t
+ipsec_inbound_cqp_dequeue(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
+		uint16_t len);
+
+uint16_t
+ipsec_outbound_cqp_dequeue(struct ipsec_ctx *ctx, struct rte_mbuf *pkts[],
+		uint16_t len);
+
 static inline uint16_t
 ipsec_metadata_size(void)
 {
@@ -248,4 +256,7 @@ sa_check_offloads(uint16_t port_id, uint64_t *rx_offloads,
 int
 add_dst_ethaddr(uint16_t port, const struct ether_addr *addr);
 
+void
+enqueue_cop_burst(struct cdev_qp *cqp);
+
 #endif /* __IPSEC_H__ */
-- 
2.17.1