[dpdk-dev] [RFC 05/12] examples/l3fwd: add support for s390x architecture

[Vivian Kong <vivkong@gmail.com>]

Add s390x specific implementation.

Signed-off-by: Vivian Kong <vivkong@ca.ibm.com>
---
 examples/l3fwd/l3fwd_em.c        |   8 +
 examples/l3fwd/l3fwd_lpm_s390x.h | 137 ++++++++++++++++
 examples/l3fwd/l3fwd_s390x.h     | 259 +++++++++++++++++++++++++++++++
 3 files changed, 404 insertions(+)
 create mode 100644 examples/l3fwd/l3fwd_lpm_s390x.h
 create mode 100644 examples/l3fwd/l3fwd_s390x.h

diff --git a/examples/l3fwd/l3fwd_em.c b/examples/l3fwd/l3fwd_em.c
index fa8f82be6..a486a9c57 100644
--- a/examples/l3fwd/l3fwd_em.c
+++ b/examples/l3fwd/l3fwd_em.c
@@ -240,6 +240,14 @@ em_mask_key(void *key, xmm_t mask)
 
 	return vec_and(data, mask);
 }
+#elif defined(RTE_MACHINE_CPUFLAG_ZARCH)
+static inline xmm_t
+em_mask_key(void *key, xmm_t mask)
+{
+	xmm_t data = (xmm_t) vec_xld2(0, (unsigned int *)(key));
+
+	return data + mask;
+}
 #else
 #error No vector engine (SSE, NEON, ALTIVEC) available, check your toolchain
 #endif
diff --git a/examples/l3fwd/l3fwd_lpm_s390x.h b/examples/l3fwd/l3fwd_lpm_s390x.h
new file mode 100644
index 000000000..858f696ba
--- /dev/null
+++ b/examples/l3fwd/l3fwd_lpm_s390x.h
@@ -0,0 +1,137 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2010-2016 Intel Corporation.
+ * (c) Copyright IBM Corp. 2017, 2019
+ */
+#ifndef __L3FWD_LPM_S390X_H__
+#define __L3FWD_LPM_S390X_H__
+
+#include "l3fwd_s390x.h"
+
+typedef unsigned char vector_unsigned_char
+	__attribute__((vector_size(16*sizeof(unsigned char))));
+
+/*
+ * Read packet_type and destination IPV4 addresses from 4 mbufs.
+ */
+static inline void
+processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
+		vector_unsigned_int *dip,
+		uint32_t *ipv4_flag)
+{
+	struct ipv4_hdr *ipv4_hdr;
+	struct ether_hdr *eth_hdr;
+	uint32_t x0, x1, x2, x3;
+
+	eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
+	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
+	x0 = ipv4_hdr->dst_addr;
+	ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
+
+	rte_compiler_barrier();
+	eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
+	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
+	x1 = ipv4_hdr->dst_addr;
+	ipv4_flag[0] &= pkt[1]->packet_type;
+
+	rte_compiler_barrier();
+	eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
+	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
+	x2 = ipv4_hdr->dst_addr;
+	ipv4_flag[0] &= pkt[2]->packet_type;
+
+	rte_compiler_barrier();
+	eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
+	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
+	x3 = ipv4_hdr->dst_addr;
+	ipv4_flag[0] &= pkt[3]->packet_type;
+
+	rte_compiler_barrier();
+	dip[0] = (vector_unsigned_int){x0, x1, x2, x3};
+}
+
+/*
+ * Lookup into LPM for destination port.
+ * If lookup fails, use incoming port (portid) as destination port.
+ */
+static inline void
+processx4_step2(const struct lcore_conf *qconf,
+		vector_unsigned_int dip,
+		uint32_t ipv4_flag,
+		uint8_t portid,
+		struct rte_mbuf *pkt[FWDSTEP],
+		uint16_t dprt[FWDSTEP])
+{
+	rte_xmm_t dst;
+	const vector_unsigned_char bswap_mask = (vector_unsigned_char){
+							3, 2, 1, 0,
+							7, 6, 5, 4,
+							11, 10, 9, 8,
+							15, 14, 13, 12};
+
+	/* Byte swap 4 IPV4 addresses. */
+	dip = (vector_unsigned_int)vec_perm(*(vector_unsigned_char *)&dip,
+					(vector_unsigned_char){}, bswap_mask);
+
+	/* if all 4 packets are IPV4. */
+	if (likely(ipv4_flag)) {
+		rte_lpm_lookupx4(qconf->ipv4_lookup_struct, (xmm_t)dip,
+			(uint32_t *)&dst, portid);
+		/* get rid of unused upper 16 bit for each dport. */
+		dst.x = (xmm_t)vec_packs(dst.x, dst.x);
+		*(uint64_t *)dprt = dst.u64[0];
+	} else {
+		dst.x = (xmm_t)dip;
+		dprt[0] = lpm_get_dst_port_with_ipv4(qconf, pkt[0],
+							dst.u32[0], portid);
+		dprt[1] = lpm_get_dst_port_with_ipv4(qconf, pkt[1],
+							dst.u32[1], portid);
+		dprt[2] = lpm_get_dst_port_with_ipv4(qconf, pkt[2],
+							dst.u32[2], portid);
+		dprt[3] = lpm_get_dst_port_with_ipv4(qconf, pkt[3],
+							dst.u32[3], portid);
+	}
+}
+
+/*
+ * Buffer optimized handling of packets, invoked
+ * from main_loop.
+ */
+static inline void
+l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
+			uint8_t portid, struct lcore_conf *qconf)
+{
+	int32_t j;
+	uint16_t dst_port[MAX_PKT_BURST];
+	vector_unsigned_int dip[MAX_PKT_BURST / FWDSTEP];
+	uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
+	const int32_t k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
+
+	for (j = 0; j != k; j += FWDSTEP)
+		processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
+				&ipv4_flag[j / FWDSTEP]);
+
+	for (j = 0; j != k; j += FWDSTEP)
+		processx4_step2(qconf, dip[j / FWDSTEP],
+				ipv4_flag[j / FWDSTEP],
+				portid, &pkts_burst[j], &dst_port[j]);
+
+	/* Classify last up to 3 packets one by one */
+	switch (nb_rx % FWDSTEP) {
+	case 3:
+		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
+		j++;
+		/* fall-through */
+	case 2:
+		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
+		j++;
+		/* fall-through */
+	case 1:
+		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
+		j++;
+		/* fall-through */
+	}
+
+	send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
+}
+
+#endif /* __L3FWD_LPM_S390X_H__ */
diff --git a/examples/l3fwd/l3fwd_s390x.h b/examples/l3fwd/l3fwd_s390x.h
new file mode 100644
index 000000000..d027092a4
--- /dev/null
+++ b/examples/l3fwd/l3fwd_s390x.h
@@ -0,0 +1,259 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2016 Intel Corporation.
+ * (c) Copyright IBM Corp. 2017, 2019
+ */
+#ifndef _L3FWD_S390X_H_
+#define _L3FWD_S390X_H_
+
+#include "l3fwd.h"
+#include "l3fwd_common.h"
+
+#define vec_sro(a, b) vec_srb(a, (b) << 64) // Vector Shift Right by Octet
+typedef unsigned int vector_unsigned_int
+	__attribute__((vector_size(4*sizeof(unsigned int))));
+typedef unsigned short vector_unsigned_short
+	__attribute__((vector_size(8*sizeof(unsigned short))));
+
+/*
+ * Update source and destination MAC addresses in the ethernet header.
+ * Perform RFC1812 checks and updates for IPV4 packets.
+ */
+static inline void
+processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
+{
+	vector_unsigned_int te[FWDSTEP];
+	vector_unsigned_int ve[FWDSTEP];
+	vector_unsigned_int *p[FWDSTEP];
+
+	p[0] = rte_pktmbuf_mtod(pkt[0], vector_unsigned_int *);
+	p[1] = rte_pktmbuf_mtod(pkt[1], vector_unsigned_int *);
+	p[2] = rte_pktmbuf_mtod(pkt[2], vector_unsigned_int *);
+	p[3] = rte_pktmbuf_mtod(pkt[3], vector_unsigned_int *);
+
+	ve[0] = (vector_unsigned_int)val_eth[dst_port[0]];
+	te[0] = *p[0];
+
+	ve[1] = (vector_unsigned_int)val_eth[dst_port[1]];
+	te[1] = *p[1];
+
+	ve[2] = (vector_unsigned_int)val_eth[dst_port[2]];
+	te[2] = *p[2];
+
+	ve[3] = (vector_unsigned_int)val_eth[dst_port[3]];
+	te[3] = *p[3];
+
+	/* Update first 12 bytes, keep rest bytes intact. */
+	te[0] = (vector_unsigned_int)vec_sel(
+			(vector_unsigned_short)ve[0],
+			(vector_unsigned_short)te[0],
+			(vector_unsigned_short) {0, 0, 0, 0,
+						0, 0, 0xffff, 0xffff});
+
+	te[1] = (vector_unsigned_int)vec_sel(
+			(vector_unsigned_short)ve[1],
+			(vector_unsigned_short)te[1],
+			(vector_unsigned_short) {0, 0, 0, 0,
+						0, 0, 0xffff, 0xffff});
+
+	te[2] = (vector_unsigned_int)vec_sel(
+			(vector_unsigned_short)ve[2],
+			(vector_unsigned_short)te[2],
+			(vector_unsigned_short) {0, 0, 0, 0, 0,
+						0, 0xffff, 0xffff});
+
+	te[3] = (vector_unsigned_int)vec_sel(
+			(vector_unsigned_short)ve[3],
+			(vector_unsigned_short)te[3],
+			(vector_unsigned_short) {0, 0, 0, 0,
+						0, 0, 0xffff, 0xffff});
+
+	*p[0] = te[0];
+	*p[1] = te[1];
+	*p[2] = te[2];
+	*p[3] = te[3];
+
+	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
+		&dst_port[0], pkt[0]->packet_type);
+	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
+		&dst_port[1], pkt[1]->packet_type);
+	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
+		&dst_port[2], pkt[2]->packet_type);
+	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
+		&dst_port[3], pkt[3]->packet_type);
+}
+
+/*
+ * Group consecutive packets with the same destination port in bursts of 4.
+ * Suppose we have array of destination ports:
+ * dst_port[] = {a, b, c, d,, e, ... }
+ * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
+ * We doing 4 comparisons at once and the result is 4 bit mask.
+ * This mask is used as an index into prebuild array of pnum values.
+ */
+static inline uint16_t *
+port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, vector_unsigned_short dp1,
+	vector_unsigned_short dp2)
+{
+	union {
+		uint16_t u16[FWDSTEP + 1];
+		uint64_t u64;
+	} *pnum = (void *)pn;
+
+	int32_t v;
+
+	v = vec_any_eq(dp1, dp2);
+
+
+	/* update last port counter. */
+	lp[0] += gptbl[v].lpv;
+
+	/* if dest port value has changed. */
+	if (v != GRPMSK) {
+		pnum->u64 = gptbl[v].pnum;
+		pnum->u16[FWDSTEP] = 1;
+		lp = pnum->u16 + gptbl[v].idx;
+	}
+
+	return lp;
+}
+
+/**
+ * Process one packet:
+ * Update source and destination MAC addresses in the ethernet header.
+ * Perform RFC1812 checks and updates for IPV4 packets.
+ */
+static inline void
+process_packet(struct rte_mbuf *pkt, uint16_t *dst_port)
+{
+	struct ether_hdr *eth_hdr;
+	vector_unsigned_int te, ve;
+
+	eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
+
+	te = *(vector_unsigned_int *)eth_hdr;
+	ve = (vector_unsigned_int)val_eth[dst_port[0]];
+
+	rfc1812_process((struct ipv4_hdr *)(eth_hdr + 1), dst_port,
+			pkt->packet_type);
+
+	/* dynamically vec_sel te and ve for MASK_ETH (0x3f) */
+	te = (vector_unsigned_int)vec_sel(
+		(vector_unsigned_short)ve,
+		(vector_unsigned_short)te,
+		(vector_unsigned_short){0, 0, 0, 0,
+					0, 0, 0xffff, 0xffff});
+
+	*(vector_unsigned_int *)eth_hdr = te;
+}
+
+/**
+ * Send packets burst from pkts_burst to the ports in dst_port array
+ */
+static __rte_always_inline void
+send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
+		uint16_t dst_port[MAX_PKT_BURST], int nb_rx)
+{
+	int32_t k;
+	int j = 0;
+	uint16_t dlp;
+	uint16_t *lp;
+	uint16_t pnum[MAX_PKT_BURST + 1];
+
+	/*
+	 * Finish packet processing and group consecutive
+	 * packets with the same destination port.
+	 */
+	k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
+	if (k != 0) {
+		vector_unsigned_short dp1, dp2;
+
+		lp = pnum;
+		lp[0] = 1;
+
+		processx4_step3(pkts_burst, dst_port);
+
+		/* dp1: <d[0], d[1], d[2], d[3], ... > */
+		dp1 = *(vector_unsigned_short *)dst_port;
+
+		for (j = FWDSTEP; j != k; j += FWDSTEP) {
+			processx4_step3(&pkts_burst[j], &dst_port[j]);
+
+			/*
+			 * dp2:
+			 * <d[j-3], d[j-2], d[j-1], d[j], ... >
+			 */
+			dp2 = *((vector_unsigned_short *)
+					&dst_port[j - FWDSTEP + 1]);
+			lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
+
+			/*
+			 * dp1:
+			 * <d[j], d[j+1], d[j+2], d[j+3], ... >
+			 */
+			dp1 = vec_sro(dp2, (vector unsigned char) {
+				0, 0, 0, 0, 0, 0, 0, 0,
+				0, 0, 0, (FWDSTEP - 1) * sizeof(dst_port[0])});
+		}
+
+		/*
+		 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
+		 */
+		dp2 = vec_perm(dp1, (vector_unsigned_short){},
+				(vector unsigned char) {0xf9});
+		lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
+
+		/*
+		 * remove values added by the last repeated
+		 * dst port.
+		 */
+		lp[0]--;
+		dlp = dst_port[j - 1];
+	} else {
+		/* set dlp and lp to the never used values. */
+		dlp = BAD_PORT - 1;
+		lp = pnum + MAX_PKT_BURST;
+	}
+
+	/* Process up to last 3 packets one by one. */
+	switch (nb_rx % FWDSTEP) {
+	case 3:
+		process_packet(pkts_burst[j], dst_port + j);
+		GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
+		j++;
+		/* fall-through */
+	case 2:
+		process_packet(pkts_burst[j], dst_port + j);
+		GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
+		j++;
+		/* fall-through */
+	case 1:
+		process_packet(pkts_burst[j], dst_port + j);
+		GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
+		j++;
+	}
+
+	/*
+	 * Send packets out, through destination port.
+	 * Consecutive packets with the same destination port
+	 * are already grouped together.
+	 * If destination port for the packet equals BAD_PORT,
+	 * then free the packet without sending it out.
+	 */
+	for (j = 0; j < nb_rx; j += k) {
+
+		int32_t m;
+		uint16_t pn;
+
+		pn = dst_port[j];
+		k = pnum[j];
+
+		if (likely(pn != BAD_PORT))
+			send_packetsx4(qconf, pn, pkts_burst + j, k);
+		else
+			for (m = j; m != j + k; m++)
+				rte_pktmbuf_free(pkts_burst[m]);
+
+	}
+}
+
+#endif /* _L3FWD_S390X_H_ */
-- 
2.17.1