[dpdk-dev] [PATCH v3] net/memif: zero-copy slave

Jakub Grajciar jgrajcia at cisco.com
Wed Jul 3 12:44:08 CEST 2019


Zero-copy slave support for memif PMD.
Slave interface exposes DPDK memory to
master interface. Only single file segments
are supported (EAL option --single-file-segments).

Signed-off-by: Jakub Grajciar <jgrajcia at cisco.com>
---
 doc/guides/nics/memif.rst         |  29 ++
 drivers/net/memif/Makefile        |   1 +
 drivers/net/memif/memif_rxtx.c    | 567 ++++++++++++++++++++++++++++
 drivers/net/memif/memif_rxtx.h    |  36 ++
 drivers/net/memif/memif_socket.c  |  63 ++--
 drivers/net/memif/meson.build     |   3 +-
 drivers/net/memif/rte_eth_memif.c | 594 +++++++++++-------------------
 drivers/net/memif/rte_eth_memif.h |  12 +-
 8 files changed, 887 insertions(+), 418 deletions(-)
 create mode 100644 drivers/net/memif/memif_rxtx.c
 create mode 100644 drivers/net/memif/memif_rxtx.h

V2:
- fix coding style

V3:
- fix compilation issues

diff --git a/doc/guides/nics/memif.rst b/doc/guides/nics/memif.rst
index de2d481eb..46cadb13f 100644
--- a/doc/guides/nics/memif.rst
+++ b/doc/guides/nics/memif.rst
@@ -171,6 +171,35 @@ Files
 - net/memif/memif.h *- descriptor and ring definitions*
 - net/memif/rte_eth_memif.c *- eth_memif_rx() eth_memif_tx()*

+Zero-copy slave
+~~~~~~~~~~~~~~~
+
+**Shared memory format**
+
+Region 0 is created by memif driver and contains rings. Slave interface exposes DPDK memory (memseg).
+Instead of using memfd_create() to create new shared file, existing memsegs are used.
+Master interface functions the same as with zero-copy disabled.
+
+region 0:
+
++-----------------------+
+| Rings                 |
++-----------+-----------+
+| S2M rings | M2S rings |
++-----------+-----------+
+
+region n:
+
++-----------------+
+| Buffers         |
++-----------------+
+|memseg           |
++-----------------+
+
+Buffers are dequeued and enqueued as needed. Offset descriptor field is calculated at tx.
+Only single file segments mode (EAL option --single-file-segments) is supported, as calculating
+offset from multiple segments is too expensive.
+
 Example: testpmd
 ----------------------------
 In this example we run two instances of testpmd application and transmit packets over memif.
diff --git a/drivers/net/memif/Makefile b/drivers/net/memif/Makefile
index fdbdf3378..ceb24df73 100644
--- a/drivers/net/memif/Makefile
+++ b/drivers/net/memif/Makefile
@@ -30,5 +30,6 @@ LDLIBS += -lrte_bus_vdev
 #
 SRCS-$(CONFIG_RTE_LIBRTE_PMD_MEMIF) += rte_eth_memif.c
 SRCS-$(CONFIG_RTE_LIBRTE_PMD_MEMIF) += memif_socket.c
+SRCS-$(CONFIG_RTE_LIBRTE_PMD_MEMIF) += memif_rxtx.c

 include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/drivers/net/memif/memif_rxtx.c b/drivers/net/memif/memif_rxtx.c
new file mode 100644
index 000000000..f34cafacd
--- /dev/null
+++ b/drivers/net/memif/memif_rxtx.c
@@ -0,0 +1,567 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2019 Cisco Systems, Inc.  All rights reserved.
+ */
+
+#include <unistd.h>
+#include <errno.h>
+
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_eal_memconfig.h>
+#include <rte_ethdev_vdev.h>
+#include <rte_mbuf.h>
+
+#include <rte_prefetch.h>
+
+#include "rte_eth_memif.h"
+#include "memif_rxtx.h"
+
+static void *
+memif_get_buffer(struct pmd_process_private *proc_private, memif_desc_t *d)
+{
+	return ((uint8_t *)proc_private->regions[d->region]->addr + d->offset);
+}
+
+/* Free mbufs received by master */
+static void
+memif_free_stored_mbufs(struct pmd_process_private *proc_private, struct memif_queue *mq)
+{
+	uint16_t mask = (1 << mq->log2_ring_size) - 1;
+	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
+
+	/* FIXME: improve performance */
+	while (mq->last_tail != ring->tail) {
+		RTE_MBUF_PREFETCH_TO_FREE(mq->buffers[(mq->last_tail + 1) & mask]);
+		/* Decrement refcnt and free mbuf. (current segment) */
+		rte_mbuf_refcnt_update(mq->buffers[mq->last_tail & mask], -1);
+		rte_pktmbuf_free_seg(mq->buffers[mq->last_tail & mask]);
+		mq->last_tail++;
+	}
+}
+
+static int
+memif_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *cur_tail,
+		    struct rte_mbuf *tail)
+{
+	/* Check for number-of-segments-overflow */
+	if (unlikely(head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS))
+		return -EOVERFLOW;
+
+	/* Chain 'tail' onto the old tail */
+	cur_tail->next = tail;
+
+	/* accumulate number of segments and total length. */
+	head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
+
+	tail->pkt_len = tail->data_len;
+	head->pkt_len += tail->pkt_len;
+
+	return 0;
+}
+
+uint16_t
+eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+	struct memif_queue *mq = queue;
+	struct pmd_internals *pmd = rte_eth_devices[mq->in_port].data->dev_private;
+	struct pmd_process_private *proc_private =
+		rte_eth_devices[mq->in_port].process_private;
+	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
+	uint16_t cur_slot, last_slot, n_slots, ring_size, mask, s0;
+	uint16_t n_rx_pkts = 0;
+	uint16_t mbuf_size = rte_pktmbuf_data_room_size(mq->mempool) -
+		RTE_PKTMBUF_HEADROOM;
+	uint16_t src_len, src_off, dst_len, dst_off, cp_len;
+	memif_ring_type_t type = mq->type;
+	memif_desc_t *d0;
+	struct rte_mbuf *mbuf, *mbuf_head, *mbuf_tail;
+	uint64_t b;
+	ssize_t size __rte_unused;
+	uint16_t head;
+	int ret;
+	struct rte_eth_link link;
+
+	if (unlikely((pmd->flags & ETH_MEMIF_FLAG_CONNECTED) == 0))
+		return 0;
+	if (unlikely(ring == NULL)) {
+		/* Secondary process will attempt to request regions. */
+		rte_eth_link_get(mq->in_port, &link);
+		return 0;
+	}
+
+	/* consume interrupt */
+	if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0)
+		size = read(mq->intr_handle.fd, &b, sizeof(b));
+
+	ring_size = 1 << mq->log2_ring_size;
+	mask = ring_size - 1;
+
+	cur_slot = (type == MEMIF_RING_S2M) ? mq->last_head : mq->last_tail;
+	last_slot = (type == MEMIF_RING_S2M) ? ring->head : ring->tail;
+	if (cur_slot == last_slot)
+		goto refill;
+	n_slots = last_slot - cur_slot;
+
+	while (n_slots && n_rx_pkts < nb_pkts) {
+		mbuf_head = rte_pktmbuf_alloc(mq->mempool);
+		if (unlikely(mbuf_head == NULL))
+			goto no_free_bufs;
+		mbuf = mbuf_head;
+		mbuf->port = mq->in_port;
+
+next_slot:
+		s0 = cur_slot & mask;
+		d0 = &ring->desc[s0];
+
+		src_len = d0->length;
+		dst_off = 0;
+		src_off = 0;
+
+		do {
+			dst_len = mbuf_size - dst_off;
+			if (dst_len == 0) {
+				dst_off = 0;
+				dst_len = mbuf_size;
+
+				/* store pointer to tail */
+				mbuf_tail = mbuf;
+				mbuf = rte_pktmbuf_alloc(mq->mempool);
+				if (unlikely(mbuf == NULL))
+					goto no_free_bufs;
+				mbuf->port = mq->in_port;
+				ret = memif_pktmbuf_chain(mbuf_head, mbuf_tail, mbuf);
+				if (unlikely(ret < 0)) {
+					MIF_LOG(ERR, "number-of-segments-overflow");
+					rte_pktmbuf_free(mbuf);
+					goto no_free_bufs;
+				}
+			}
+			cp_len = RTE_MIN(dst_len, src_len);
+
+			rte_pktmbuf_data_len(mbuf) += cp_len;
+			rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf);
+			if (mbuf != mbuf_head)
+				rte_pktmbuf_pkt_len(mbuf_head) += cp_len;
+
+			memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off),
+			       (uint8_t *)memif_get_buffer(proc_private, d0) + src_off,
+			       cp_len);
+
+			src_off += cp_len;
+			dst_off += cp_len;
+			src_len -= cp_len;
+		} while (src_len);
+
+		cur_slot++;
+		n_slots--;
+
+		if (d0->flags & MEMIF_DESC_FLAG_NEXT)
+			goto next_slot;
+
+		mq->n_bytes += rte_pktmbuf_pkt_len(mbuf_head);
+		*bufs++ = mbuf_head;
+		n_rx_pkts++;
+	}
+
+no_free_bufs:
+	if (type == MEMIF_RING_S2M) {
+		rte_mb();
+		ring->tail = cur_slot;
+		mq->last_head = cur_slot;
+	} else {
+		mq->last_tail = cur_slot;
+	}
+
+refill:
+	if (type == MEMIF_RING_M2S) {
+		head = ring->head;
+		n_slots = ring_size - head + mq->last_tail;
+
+		while (n_slots--) {
+			s0 = head++ & mask;
+			d0 = &ring->desc[s0];
+			d0->length = pmd->run.pkt_buffer_size;
+		}
+		rte_mb();
+		ring->head = head;
+	}
+
+	mq->n_pkts += n_rx_pkts;
+	return n_rx_pkts;
+}
+
+uint16_t
+eth_memif_rx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+	struct memif_queue *mq = queue;
+	struct pmd_internals *pmd = rte_eth_devices[mq->in_port].data->dev_private;
+	struct pmd_process_private *proc_private =
+		rte_eth_devices[mq->in_port].process_private;
+	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
+	uint16_t cur_slot, last_slot, n_slots, ring_size, mask, s0, head;
+	uint16_t n_rx_pkts = 0;
+	memif_desc_t *d0;
+	struct rte_mbuf *mbuf, *mbuf_tail;
+	struct rte_mbuf *mbuf_head = NULL;
+	int ret;
+	struct rte_eth_link link;
+
+	if (unlikely((pmd->flags & ETH_MEMIF_FLAG_CONNECTED) == 0))
+		return 0;
+	if (unlikely(ring == NULL)) {
+		/* Secondary process will attempt to request regions. */
+		rte_eth_link_get(mq->in_port, &link);
+		return 0;
+	}
+
+	/* consume interrupt */
+	if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0) {
+		uint64_t b;
+		ssize_t size __rte_unused;
+		size = read(mq->intr_handle.fd, &b, sizeof(b));
+	}
+
+	ring_size = 1 << mq->log2_ring_size;
+	mask = ring_size - 1;
+
+	cur_slot = mq->last_tail;
+	last_slot = ring->tail;
+	if (cur_slot == last_slot)
+		goto refill;
+	n_slots = last_slot - cur_slot;
+
+	while (n_slots && n_rx_pkts < nb_pkts) {
+		s0 = cur_slot & mask;
+
+		d0 = &ring->desc[s0];
+		mbuf_head = mq->buffers[s0];
+		mbuf = mbuf_head;
+
+next_slot:
+		/* prefetch next descriptor */
+		if (n_rx_pkts + 1 < nb_pkts)
+			rte_prefetch0(&ring->desc[(cur_slot + 1) & mask]);
+
+		mbuf->port = mq->in_port;
+		rte_pktmbuf_data_len(mbuf) = d0->length;
+		rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf);
+
+		mq->n_bytes += rte_pktmbuf_data_len(mbuf);
+
+		cur_slot++;
+		n_slots--;
+		if (d0->flags & MEMIF_DESC_FLAG_NEXT) {
+			s0 = cur_slot & mask;
+			d0 = &ring->desc[s0];
+			mbuf_tail = mbuf;
+			mbuf = mq->buffers[s0];
+			ret = memif_pktmbuf_chain(mbuf_head, mbuf_tail, mbuf);
+			if (unlikely(ret < 0)) {
+				MIF_LOG(ERR, "number-of-segments-overflow");
+				goto refill;
+			}
+			goto next_slot;
+		}
+
+		*bufs++ = mbuf_head;
+		n_rx_pkts++;
+	}
+
+	mq->last_tail = cur_slot;
+
+/* Supply master with new buffers */
+refill:
+	head = ring->head;
+	n_slots = ring_size - head + mq->last_tail;
+
+	if (n_slots < 32)
+		goto no_free_mbufs;
+
+	ret = rte_pktmbuf_alloc_bulk(mq->mempool, &mq->buffers[head & mask], n_slots);
+	if (unlikely(ret < 0))
+		goto no_free_mbufs;
+
+	while (n_slots--) {
+		s0 = head++ & mask;
+		if (n_slots > 0)
+			rte_prefetch0(mq->buffers[head & mask]);
+		d0 = &ring->desc[s0];
+		/* store buffer header */
+		mbuf = mq->buffers[s0];
+		/* populate descriptor */
+		d0->length = rte_pktmbuf_data_room_size(mq->mempool) -
+				RTE_PKTMBUF_HEADROOM;
+		d0->region = 1;
+		d0->offset = rte_pktmbuf_mtod(mbuf, uint8_t *) -
+			(uint8_t *)proc_private->regions[d0->region]->addr;
+	}
+no_free_mbufs:
+	rte_mb();
+	ring->head = head;
+
+	mq->n_pkts += n_rx_pkts;
+
+	return n_rx_pkts;
+}
+
+uint16_t
+eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+	struct memif_queue *mq = queue;
+	struct pmd_internals *pmd = rte_eth_devices[mq->in_port].data->dev_private;
+	struct pmd_process_private *proc_private =
+		rte_eth_devices[mq->in_port].process_private;
+	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
+	uint16_t slot, saved_slot, n_free, ring_size, mask, n_tx_pkts = 0;
+	uint16_t src_len, src_off, dst_len, dst_off, cp_len;
+	memif_ring_type_t type = mq->type;
+	memif_desc_t *d0;
+	struct rte_mbuf *mbuf;
+	struct rte_mbuf *mbuf_head;
+	uint64_t a;
+	ssize_t size;
+	struct rte_eth_link link;
+
+	if (unlikely((pmd->flags & ETH_MEMIF_FLAG_CONNECTED) == 0))
+		return 0;
+	if (unlikely(ring == NULL)) {
+		/* Secondary process will attempt to request regions. */
+		rte_eth_link_get(mq->in_port, &link);
+		return 0;
+	}
+
+	ring_size = 1 << mq->log2_ring_size;
+	mask = ring_size - 1;
+
+	n_free = ring->tail - mq->last_tail;
+	mq->last_tail += n_free;
+	slot = (type == MEMIF_RING_S2M) ? ring->head : ring->tail;
+
+	if (type == MEMIF_RING_S2M)
+		n_free = ring_size - ring->head + mq->last_tail;
+	else
+		n_free = ring->head - ring->tail;
+
+	while (n_tx_pkts < nb_pkts && n_free) {
+		mbuf_head = *bufs++;
+		mbuf = mbuf_head;
+
+		saved_slot = slot;
+		d0 = &ring->desc[slot & mask];
+		dst_off = 0;
+		dst_len = (type == MEMIF_RING_S2M) ?
+			pmd->run.pkt_buffer_size : d0->length;
+
+next_in_chain:
+		src_off = 0;
+		src_len = rte_pktmbuf_data_len(mbuf);
+
+		while (src_len) {
+			if (dst_len == 0) {
+				if (n_free) {
+					slot++;
+					n_free--;
+					d0->flags |= MEMIF_DESC_FLAG_NEXT;
+					d0 = &ring->desc[slot & mask];
+					dst_off = 0;
+					dst_len = (type == MEMIF_RING_S2M) ?
+					    pmd->run.pkt_buffer_size : d0->length;
+					d0->flags = 0;
+				} else {
+					slot = saved_slot;
+					goto no_free_slots;
+				}
+			}
+			cp_len = RTE_MIN(dst_len, src_len);
+
+			memcpy((uint8_t *)memif_get_buffer(proc_private, d0) + dst_off,
+			       rte_pktmbuf_mtod_offset(mbuf, void *, src_off),
+			       cp_len);
+
+			mq->n_bytes += cp_len;
+			src_off += cp_len;
+			dst_off += cp_len;
+			src_len -= cp_len;
+			dst_len -= cp_len;
+
+			d0->length = dst_off;
+		}
+
+		if (rte_pktmbuf_is_contiguous(mbuf) == 0) {
+			mbuf = mbuf->next;
+			goto next_in_chain;
+		}
+
+		n_tx_pkts++;
+		slot++;
+		n_free--;
+		rte_pktmbuf_free(mbuf_head);
+	}
+
+no_free_slots:
+	rte_mb();
+	if (type == MEMIF_RING_S2M)
+		ring->head = slot;
+	else
+		ring->tail = slot;
+
+	if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0) {
+		a = 1;
+		size = write(mq->intr_handle.fd, &a, sizeof(a));
+		if (unlikely(size < 0)) {
+			MIF_LOG(WARNING,
+				"Failed to send interrupt. %s", strerror(errno));
+		}
+	}
+
+	mq->n_err += nb_pkts - n_tx_pkts;
+	mq->n_pkts += n_tx_pkts;
+	return n_tx_pkts;
+}
+
+static inline int
+memif_tx_one_zc(struct pmd_process_private *proc_private, struct memif_queue *mq,
+		memif_ring_t *ring, struct rte_mbuf *mbuf, const uint16_t mask,
+		uint16_t slot, uint16_t n_free)
+{
+	memif_desc_t *d0;
+	int used_slots = 1;
+
+next_in_chain:
+	/* store pointer to mbuf to free it later */
+	mq->buffers[slot & mask] = mbuf;
+	/* Increment refcnt to make sure the buffer is not freed before master
+	 * receives it. (current segment)
+	 */
+	rte_mbuf_refcnt_update(mbuf, 1);
+	/* populate descriptor */
+	d0 = &ring->desc[slot & mask];
+	d0->length = rte_pktmbuf_data_len(mbuf);
+	/* FIXME: get region index */
+	d0->region = 1;
+	d0->offset = rte_pktmbuf_mtod(mbuf, uint8_t *) -
+		(uint8_t *)proc_private->regions[d0->region]->addr;
+	d0->flags = 0;
+
+	/* check if buffer is chained */
+	if (rte_pktmbuf_is_contiguous(mbuf) == 0) {
+		if (n_free < 2)
+			return 0;
+		/* mark buffer as chained */
+		d0->flags |= MEMIF_DESC_FLAG_NEXT;
+		/* advance mbuf */
+		mbuf = mbuf->next;
+		/* update counters */
+		used_slots++;
+		slot++;
+		n_free--;
+		goto next_in_chain;
+	}
+	return used_slots;
+}
+
+uint16_t
+eth_memif_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+	struct memif_queue *mq = queue;
+	struct pmd_internals *pmd = rte_eth_devices[mq->in_port].data->dev_private;
+	struct pmd_process_private *proc_private =
+		rte_eth_devices[mq->in_port].process_private;
+	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
+	uint16_t slot, n_free, ring_size, mask, n_tx_pkts = 0;
+	memif_ring_type_t type = mq->type;
+	struct rte_eth_link link;
+
+	if (unlikely((pmd->flags & ETH_MEMIF_FLAG_CONNECTED) == 0))
+		return 0;
+	if (unlikely(ring == NULL)) {
+		/* Secondary process will attempt to request regions. */
+		rte_eth_link_get(mq->in_port, &link);
+		return 0;
+	}
+
+	ring_size = 1 << mq->log2_ring_size;
+	mask = ring_size - 1;
+
+	/* free mbufs received by master */
+	memif_free_stored_mbufs(proc_private, mq);
+
+	/* ring type always MEMIF_RING_S2M */
+	slot = ring->head;
+	n_free = ring_size - ring->head + mq->last_tail;
+
+	int used_slots;
+
+	while (n_free && (n_tx_pkts < nb_pkts)) {
+		while ((n_free > 4) && ((nb_pkts - n_tx_pkts) > 4)) {
+			if ((nb_pkts - n_tx_pkts) > 8) {
+				rte_prefetch0(*bufs + 4);
+				rte_prefetch0(*bufs + 5);
+				rte_prefetch0(*bufs + 6);
+				rte_prefetch0(*bufs + 7);
+			}
+			used_slots = memif_tx_one_zc(proc_private, mq, ring, *bufs++,
+				mask, slot, n_free);
+			if (unlikely(used_slots < 1))
+				goto no_free_slots;
+			n_tx_pkts++;
+			slot += used_slots;
+			n_free -= used_slots;
+
+			used_slots = memif_tx_one_zc(proc_private, mq, ring, *bufs++,
+				mask, slot, n_free);
+			if (unlikely(used_slots < 1))
+				goto no_free_slots;
+			n_tx_pkts++;
+			slot += used_slots;
+			n_free -= used_slots;
+
+			used_slots = memif_tx_one_zc(proc_private, mq, ring, *bufs++,
+				mask, slot, n_free);
+			if (unlikely(used_slots < 1))
+				goto no_free_slots;
+			n_tx_pkts++;
+			slot += used_slots;
+			n_free -= used_slots;
+
+			used_slots = memif_tx_one_zc(proc_private, mq, ring, *bufs++,
+				mask, slot, n_free);
+			if (unlikely(used_slots < 1))
+				goto no_free_slots;
+			n_tx_pkts++;
+			slot += used_slots;
+			n_free -= used_slots;
+		}
+		used_slots = memif_tx_one_zc(proc_private, mq, ring, *bufs++,
+			mask, slot, n_free);
+		if (unlikely(used_slots < 1))
+			goto no_free_slots;
+		n_tx_pkts++;
+		slot += used_slots;
+		n_free -= used_slots;
+	}
+
+no_free_slots:
+	rte_mb();
+	/* update ring pointers */
+	if (type == MEMIF_RING_S2M)
+		ring->head = slot;
+	else
+		ring->tail = slot;
+
+	/* Send interrupt, if enabled. */
+	if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0) {
+		uint64_t a = 1;
+		ssize_t size = write(mq->intr_handle.fd, &a, sizeof(a));
+		if (unlikely(size < 0)) {
+			MIF_LOG(WARNING,
+				"Failed to send interrupt. %s", strerror(errno));
+		}
+	}
+
+	/* increment queue counters */
+	mq->n_err += nb_pkts - n_tx_pkts;
+	mq->n_pkts += n_tx_pkts;
+
+	return n_tx_pkts;
+}
diff --git a/drivers/net/memif/memif_rxtx.h b/drivers/net/memif/memif_rxtx.h
new file mode 100644
index 000000000..1d00865b8
--- /dev/null
+++ b/drivers/net/memif/memif_rxtx.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2019 Cisco Systems, Inc.  All rights reserved.
+ */
+
+#ifndef _MEMIF_RX_TX_H_
+#define _MEMIF_RX_TX_H_
+
+#include "memif.h"
+
+/**
+ * Ger memif ring from shared memory.
+ *
+ * @param pmd
+ *   device internals
+ * @param type
+ *   memif ring direction
+ * @param ring_idx
+ *   ring index
+ *
+ * @return
+ *   - memif ring
+ */
+
+uint16_t
+eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts);
+
+uint16_t
+eth_memif_rx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts);
+
+uint16_t
+eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts);
+
+uint16_t
+eth_memif_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts);
+
+#endif /* MEMIF_RX_TX_H */
diff --git a/drivers/net/memif/memif_socket.c b/drivers/net/memif/memif_socket.c
index 01a935f87..2bd7f2927 100644
--- a/drivers/net/memif/memif_socket.c
+++ b/drivers/net/memif/memif_socket.c
@@ -176,8 +176,7 @@ memif_msg_receive_hello(struct rte_eth_dev *dev, memif_msg_t *msg)

 	strlcpy(pmd->remote_name, (char *)h->name, sizeof(pmd->remote_name));

-	MIF_LOG(DEBUG, "%s: Connecting to %s.",
-		rte_vdev_device_name(pmd->vdev), pmd->remote_name);
+	MIF_LOG(DEBUG, "Connecting to %s.", pmd->remote_name);

 	return 0;
 }
@@ -339,8 +338,7 @@ memif_msg_receive_connect(struct rte_eth_dev *dev, memif_msg_t *msg)

 	strlcpy(pmd->remote_if_name, (char *)c->if_name,
 		sizeof(pmd->remote_if_name));
-	MIF_LOG(INFO, "%s: Remote interface %s connected.",
-		rte_vdev_device_name(pmd->vdev), pmd->remote_if_name);
+	MIF_LOG(INFO, "Remote interface %s connected.", pmd->remote_if_name);

 	return 0;
 }
@@ -358,8 +356,7 @@ memif_msg_receive_connected(struct rte_eth_dev *dev, memif_msg_t *msg)

 	strlcpy(pmd->remote_if_name, (char *)c->if_name,
 		sizeof(pmd->remote_if_name));
-	MIF_LOG(INFO, "%s: Remote interface %s connected.",
-		rte_vdev_device_name(pmd->vdev), pmd->remote_if_name);
+	MIF_LOG(INFO, "Remote interface %s connected.", pmd->remote_if_name);

 	return 0;
 }
@@ -370,14 +367,13 @@ memif_msg_receive_disconnect(struct rte_eth_dev *dev, memif_msg_t *msg)
 	struct pmd_internals *pmd = dev->data->dev_private;
 	memif_msg_disconnect_t *d = &msg->disconnect;

-	memset(pmd->remote_disc_string, 0, ETH_MEMIF_DISC_STRING_SIZE);
+	memset(pmd->remote_disc_string, 0, sizeof(pmd->remote_disc_string));
 	strlcpy(pmd->remote_disc_string, (char *)d->string,
 		sizeof(pmd->remote_disc_string));

-	MIF_LOG(INFO, "%s: Disconnect received: %s",
-		rte_vdev_device_name(pmd->vdev), pmd->remote_disc_string);
+	MIF_LOG(INFO, "Disconnect received: %s", pmd->remote_disc_string);

-	memset(pmd->local_disc_string, 0, ETH_MEMIF_DISC_STRING_SIZE);
+	memset(pmd->local_disc_string, 0, 96);
 	memif_disconnect(dev);
 	return 0;
 }
@@ -473,7 +469,6 @@ memif_msg_enq_connect(struct rte_eth_dev *dev)
 {
 	struct pmd_internals *pmd = dev->data->dev_private;
 	struct memif_msg_queue_elt *e = memif_msg_enq(pmd->cc);
-	const char *name = rte_vdev_device_name(pmd->vdev);
 	memif_msg_connect_t *c;

 	if (e == NULL)
@@ -481,7 +476,7 @@ memif_msg_enq_connect(struct rte_eth_dev *dev)

 	c = &e->msg.connect;
 	e->msg.type = MEMIF_MSG_TYPE_CONNECT;
-	strlcpy((char *)c->if_name, name, sizeof(c->if_name));
+	strlcpy((char *)c->if_name, dev->data->name, sizeof(c->if_name));

 	return 0;
 }
@@ -491,7 +486,6 @@ memif_msg_enq_connected(struct rte_eth_dev *dev)
 {
 	struct pmd_internals *pmd = dev->data->dev_private;
 	struct memif_msg_queue_elt *e = memif_msg_enq(pmd->cc);
-	const char *name = rte_vdev_device_name(pmd->vdev);
 	memif_msg_connected_t *c;

 	if (e == NULL)
@@ -499,7 +493,7 @@ memif_msg_enq_connected(struct rte_eth_dev *dev)

 	c = &e->msg.connected;
 	e->msg.type = MEMIF_MSG_TYPE_CONNECTED;
-	strlcpy((char *)c->if_name, name, sizeof(c->if_name));
+	strlcpy((char *)c->if_name, dev->data->name, sizeof(c->if_name));

 	return 0;
 }
@@ -525,7 +519,6 @@ void
 memif_disconnect(struct rte_eth_dev *dev)
 {
 	struct pmd_internals *pmd = dev->data->dev_private;
-	struct pmd_process_private *proc_private = dev->process_private;
 	struct memif_msg_queue_elt *elt, *next;
 	struct memif_queue *mq;
 	struct rte_intr_handle *ih;
@@ -615,7 +608,7 @@ memif_disconnect(struct rte_eth_dev *dev)
 		}
 	}

-	memif_free_regions(proc_private);
+	memif_free_regions(dev);

 	/* reset connection configuration */
 	memset(&pmd->run, 0, sizeof(pmd->run));
@@ -662,7 +655,7 @@ memif_msg_receive(struct memif_control_channel *cc)
 			if (cmsg->cmsg_type == SCM_CREDENTIALS)
 				cr = (struct ucred *)CMSG_DATA(cmsg);
 			else if (cmsg->cmsg_type == SCM_RIGHTS)
-				memcpy(&afd, CMSG_DATA(cmsg), sizeof(int));
+				rte_memcpy(&afd, CMSG_DATA(cmsg), sizeof(int));
 		}
 		cmsg = CMSG_NXTHDR(&mh, cmsg);
 	}
@@ -861,7 +854,7 @@ memif_listener_handler(void *arg)
 }

 static struct memif_socket *
-memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
+memif_socket_create(char *key, uint8_t listener)
 {
 	struct memif_socket *sock;
 	struct sockaddr_un un;
@@ -899,17 +892,15 @@ memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
 		if (ret < 0)
 			goto error;

-		MIF_LOG(DEBUG, "%s: Memif listener socket %s created.",
-			rte_vdev_device_name(pmd->vdev), sock->filename);
+		MIF_LOG(DEBUG, "Memif listener socket %s created.", sock->filename);

 		sock->intr_handle.fd = sockfd;
 		sock->intr_handle.type = RTE_INTR_HANDLE_EXT;
 		ret = rte_intr_callback_register(&sock->intr_handle,
 						 memif_listener_handler, sock);
 		if (ret < 0) {
-			MIF_LOG(ERR, "%s: Failed to register interrupt "
-				"callback for listener socket",
-				rte_vdev_device_name(pmd->vdev));
+			MIF_LOG(ERR, "Failed to register interrupt "
+				"callback for listener socket");
 			return NULL;
 		}
 	}
@@ -917,8 +908,7 @@ memif_socket_create(struct pmd_internals *pmd, char *key, uint8_t listener)
 	return sock;

  error:
-	MIF_LOG(ERR, "%s: Failed to setup socket %s: %s",
-		rte_vdev_device_name(pmd->vdev), key, strerror(errno));
+	MIF_LOG(ERR, "Failed to setup socket %s: %s", key, strerror(errno));
 	if (sock != NULL)
 		rte_free(sock);
 	return NULL;
@@ -960,9 +950,8 @@ memif_socket_init(struct rte_eth_dev *dev, const char *socket_filename)
 	rte_memcpy(key, socket_filename, strlen(socket_filename));
 	ret = rte_hash_lookup_data(hash, key, (void **)&socket);
 	if (ret < 0) {
-		socket = memif_socket_create(pmd, key,
-					     (pmd->role ==
-					      MEMIF_ROLE_SLAVE) ? 0 : 1);
+		socket = memif_socket_create(key,
+					     (pmd->role == MEMIF_ROLE_SLAVE) ? 0 : 1);
 		if (socket == NULL)
 			return -1;
 		ret = rte_hash_add_key_data(hash, key, socket);
@@ -993,8 +982,7 @@ memif_socket_init(struct rte_eth_dev *dev, const char *socket_filename)

 	elt = rte_malloc("pmd-queue", sizeof(struct memif_socket_dev_list_elt), 0);
 	if (elt == NULL) {
-		MIF_LOG(ERR, "%s: Failed to add device to socket device list.",
-			rte_vdev_device_name(pmd->vdev));
+		MIF_LOG(ERR, "Failed to add device to socket device list.");
 		return -1;
 	}
 	elt->dev = dev;
@@ -1068,8 +1056,7 @@ memif_connect_slave(struct rte_eth_dev *dev)

 	sockfd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
 	if (sockfd < 0) {
-		MIF_LOG(ERR, "%s: Failed to open socket.",
-			rte_vdev_device_name(pmd->vdev));
+		MIF_LOG(ERR, "Failed to open socket.");
 		return -1;
 	}

@@ -1080,19 +1067,16 @@ memif_connect_slave(struct rte_eth_dev *dev)
 	ret = connect(sockfd, (struct sockaddr *)&sun,
 		      sizeof(struct sockaddr_un));
 	if (ret < 0) {
-		MIF_LOG(ERR, "%s: Failed to connect socket: %s.",
-			rte_vdev_device_name(pmd->vdev), pmd->socket_filename);
+		MIF_LOG(ERR, "Failed to connect socket: %s.", pmd->socket_filename);
 		goto error;
 	}

-	MIF_LOG(DEBUG, "%s: Memif socket: %s connected.",
-		rte_vdev_device_name(pmd->vdev), pmd->socket_filename);
+	MIF_LOG(DEBUG, "Memif socket: %s connected.", pmd->socket_filename);

 	pmd->cc = rte_zmalloc("memif-cc",
 			      sizeof(struct memif_control_channel), 0);
 	if (pmd->cc == NULL) {
-		MIF_LOG(ERR, "%s: Failed to allocate control channel.",
-			rte_vdev_device_name(pmd->vdev));
+		MIF_LOG(ERR, "Failed to allocate control channel.");
 		goto error;
 	}

@@ -1105,8 +1089,7 @@ memif_connect_slave(struct rte_eth_dev *dev)
 	ret = rte_intr_callback_register(&pmd->cc->intr_handle,
 					 memif_intr_handler, pmd->cc);
 	if (ret < 0) {
-		MIF_LOG(ERR, "%s: Failed to register interrupt callback "
-			"for control fd", rte_vdev_device_name(pmd->vdev));
+		MIF_LOG(ERR, "Failed to register interrupt callback for control fd");
 		goto error;
 	}

diff --git a/drivers/net/memif/meson.build b/drivers/net/memif/meson.build
index bedc97311..ce7acd1a0 100644
--- a/drivers/net/memif/meson.build
+++ b/drivers/net/memif/meson.build
@@ -6,7 +6,8 @@ if host_machine.system() != 'linux'
 endif

 sources = files('rte_eth_memif.c',
-		'memif_socket.c')
+		'memif_socket.c',
+		'memif_rxtx.c')

 allow_experimental_apis = true
 # Experimantal APIs:
diff --git a/drivers/net/memif/rte_eth_memif.c b/drivers/net/memif/rte_eth_memif.c
index e9ddf6413..324700a2f 100644
--- a/drivers/net/memif/rte_eth_memif.c
+++ b/drivers/net/memif/rte_eth_memif.c
@@ -23,9 +23,14 @@
 #include <rte_kvargs.h>
 #include <rte_bus_vdev.h>
 #include <rte_string_fns.h>
+#include <rte_errno.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_eal_memconfig.h>

 #include "rte_eth_memif.h"
 #include "memif_socket.h"
+#include "memif_rxtx.h"

 #define ETH_MEMIF_ID_ARG		"id"
 #define ETH_MEMIF_ROLE_ARG		"role"
@@ -56,6 +61,122 @@ memif_version(void)
 	return ("memif-" RTE_STR(MEMIF_VERSION_MAJOR) "." RTE_STR(MEMIF_VERSION_MINOR));
 }

+static int
+memif_region_init_zc(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
+		     void *arg)
+{
+	struct pmd_process_private *proc_private = (struct pmd_process_private *)arg;
+	struct memif_region *r;
+
+	if (proc_private->regions_num < 1) {
+		MIF_LOG(ERR, "Missing descriptor region");
+		return -1;
+	}
+
+	r = proc_private->regions[proc_private->regions_num - 1];
+
+	if (r->addr != msl->base_va)
+		r = proc_private->regions[++proc_private->regions_num - 1];
+
+	if (r == NULL) {
+		r = rte_zmalloc("region", sizeof(struct memif_region), 0);
+		if (r == NULL) {
+			MIF_LOG(ERR, "Failed to alloc memif region.");
+			return -ENOMEM;
+		}
+
+		r->addr = msl->base_va;
+		r->region_size = ms->len;
+		r->fd = rte_memseg_get_fd(ms);
+		if (r->fd < 0)
+			return -1;
+		r->pkt_buffer_offset = 0;
+
+		proc_private->regions[proc_private->regions_num - 1] = r;
+	} else {
+		r->region_size += ms->len;
+	}
+
+	return 0;
+}
+
+static int
+memif_region_init_shm(struct rte_eth_dev *dev, uint8_t has_buffers)
+{
+	struct pmd_internals *pmd = dev->data->dev_private;
+	struct pmd_process_private *proc_private = dev->process_private;
+	char shm_name[ETH_MEMIF_SHM_NAME_SIZE];
+	int ret = 0;
+	struct memif_region *r;
+
+	if (proc_private->regions_num >= ETH_MEMIF_MAX_REGION_NUM) {
+		MIF_LOG(ERR, "Too many regions.");
+		return -1;
+	}
+
+	r = rte_zmalloc("region", sizeof(struct memif_region), 0);
+	if (r == NULL) {
+		MIF_LOG(ERR, "Failed to alloc memif region.");
+		return -ENOMEM;
+	}
+
+	/* calculate buffer offset */
+	r->pkt_buffer_offset = (pmd->run.num_s2m_rings + pmd->run.num_m2s_rings) *
+	    (sizeof(memif_ring_t) + sizeof(memif_desc_t) *
+	    (1 << pmd->run.log2_ring_size));
+
+	r->region_size = r->pkt_buffer_offset;
+	/* if region has buffers, add buffers size to region_size */
+	if (has_buffers == 1)
+		r->region_size += (uint32_t)(pmd->run.pkt_buffer_size *
+			(1 << pmd->run.log2_ring_size) *
+			(pmd->run.num_s2m_rings +
+			 pmd->run.num_m2s_rings));
+
+	memset(shm_name, 0, sizeof(char) * ETH_MEMIF_SHM_NAME_SIZE);
+	snprintf(shm_name, ETH_MEMIF_SHM_NAME_SIZE, "memif_region_%d",
+		 proc_private->regions_num);
+
+	r->fd = memfd_create(shm_name, MFD_ALLOW_SEALING);
+	if (r->fd < 0) {
+		MIF_LOG(ERR, "Failed to create shm file: %s.", strerror(errno));
+		ret = -1;
+		goto error;
+	}
+
+	ret = fcntl(r->fd, F_ADD_SEALS, F_SEAL_SHRINK);
+	if (ret < 0) {
+		MIF_LOG(ERR, "Failed to add seals to shm file: %s.", strerror(errno));
+		goto error;
+	}
+
+	ret = ftruncate(r->fd, r->region_size);
+	if (ret < 0) {
+		MIF_LOG(ERR, "Failed to truncate shm file: %s.", strerror(errno));
+		goto error;
+	}
+
+	r->addr = mmap(NULL, r->region_size, PROT_READ |
+		       PROT_WRITE, MAP_SHARED, r->fd, 0);
+	if (r->addr == MAP_FAILED) {
+		MIF_LOG(ERR, "Failed to mmap shm region: %s.", strerror(ret));
+		ret = -1;
+		goto error;
+	}
+
+	proc_private->regions[proc_private->regions_num] = r;
+	proc_private->regions_num++;
+
+	return ret;
+
+error:
+	if (r->fd > 0)
+		close(r->fd);
+	r->fd = -1;
+
+	return ret;
+}
+
 /* Message header to synchronize regions */
 struct mp_region_msg {
 	char port_name[RTE_DEV_NAME_MAX_LEN];
@@ -116,10 +237,14 @@ memif_mp_request_regions(struct rte_eth_dev *dev)
 	struct mp_region_msg *reply_param;
 	struct memif_region *r;
 	struct pmd_process_private *proc_private = dev->process_private;
+	struct pmd_internals *pmd = dev->data->dev_private;
+	/* in case of zero-copy slave, only request region 0 */
+	uint16_t max_region_num = (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY) ?
+				   1 : ETH_MEMIF_MAX_REGION_NUM;

 	MIF_LOG(DEBUG, "Requesting memory regions");

-	for (i = 0; i < ETH_MEMIF_MAX_REGION_NUM; i++) {
+	for (i = 0; i < max_region_num; i++) {
 		/* Prepare the message */
 		memset(&msg, 0, sizeof(msg));
 		strlcpy(msg.name, MEMIF_MP_SEND_REGION, sizeof(msg.name));
@@ -161,6 +286,12 @@ memif_mp_request_regions(struct rte_eth_dev *dev)
 		free(reply);
 	}

+	if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY) {
+		ret = rte_memseg_walk(memif_region_init_zc, (void *)proc_private);
+		if (ret < 0)
+			return ret;
+	}
+
 	return memif_connect(dev);
 }

@@ -199,7 +330,7 @@ memif_get_ring_offset(struct rte_eth_dev *dev, struct memif_queue *mq,
 		(uint8_t *)proc_private->regions[mq->region]->addr);
 }

-static memif_ring_t *
+memif_ring_t *
 memif_get_ring_from_queue(struct pmd_process_private *proc_private,
 			  struct memif_queue *mq)
 {
@@ -212,291 +343,26 @@ memif_get_ring_from_queue(struct pmd_process_private *proc_private,
 	return (memif_ring_t *)((uint8_t *)r->addr + mq->ring_offset);
 }

-static void *
-memif_get_buffer(struct pmd_process_private *proc_private, memif_desc_t *d)
-{
-	return ((uint8_t *)proc_private->regions[d->region]->addr + d->offset);
-}
-
-static int
-memif_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *cur_tail,
-		    struct rte_mbuf *tail)
-{
-	/* Check for number-of-segments-overflow */
-	if (unlikely(head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS))
-		return -EOVERFLOW;
-
-	/* Chain 'tail' onto the old tail */
-	cur_tail->next = tail;
-
-	/* accumulate number of segments and total length. */
-	head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
-
-	tail->pkt_len = tail->data_len;
-	head->pkt_len += tail->pkt_len;
-
-	return 0;
-}
-
-static uint16_t
-eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
-{
-	struct memif_queue *mq = queue;
-	struct pmd_internals *pmd = rte_eth_devices[mq->in_port].data->dev_private;
-	struct pmd_process_private *proc_private =
-		rte_eth_devices[mq->in_port].process_private;
-	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
-	uint16_t cur_slot, last_slot, n_slots, ring_size, mask, s0;
-	uint16_t n_rx_pkts = 0;
-	uint16_t mbuf_size = rte_pktmbuf_data_room_size(mq->mempool) -
-		RTE_PKTMBUF_HEADROOM;
-	uint16_t src_len, src_off, dst_len, dst_off, cp_len;
-	memif_ring_type_t type = mq->type;
-	memif_desc_t *d0;
-	struct rte_mbuf *mbuf, *mbuf_head, *mbuf_tail;
-	uint64_t b;
-	ssize_t size __rte_unused;
-	uint16_t head;
-	int ret;
-	struct rte_eth_link link;
-
-	if (unlikely((pmd->flags & ETH_MEMIF_FLAG_CONNECTED) == 0))
-		return 0;
-	if (unlikely(ring == NULL)) {
-		/* Secondary process will attempt to request regions. */
-		rte_eth_link_get(mq->in_port, &link);
-		return 0;
-	}
-
-	/* consume interrupt */
-	if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0)
-		size = read(mq->intr_handle.fd, &b, sizeof(b));
-
-	ring_size = 1 << mq->log2_ring_size;
-	mask = ring_size - 1;
-
-	cur_slot = (type == MEMIF_RING_S2M) ? mq->last_head : mq->last_tail;
-	last_slot = (type == MEMIF_RING_S2M) ? ring->head : ring->tail;
-	if (cur_slot == last_slot)
-		goto refill;
-	n_slots = last_slot - cur_slot;
-
-	while (n_slots && n_rx_pkts < nb_pkts) {
-		mbuf_head = rte_pktmbuf_alloc(mq->mempool);
-		if (unlikely(mbuf_head == NULL))
-			goto no_free_bufs;
-		mbuf = mbuf_head;
-		mbuf->port = mq->in_port;
-
-next_slot:
-		s0 = cur_slot & mask;
-		d0 = &ring->desc[s0];
-
-		src_len = d0->length;
-		dst_off = 0;
-		src_off = 0;
-
-		do {
-			dst_len = mbuf_size - dst_off;
-			if (dst_len == 0) {
-				dst_off = 0;
-				dst_len = mbuf_size;
-
-				/* store pointer to tail */
-				mbuf_tail = mbuf;
-				mbuf = rte_pktmbuf_alloc(mq->mempool);
-				if (unlikely(mbuf == NULL))
-					goto no_free_bufs;
-				mbuf->port = mq->in_port;
-				ret = memif_pktmbuf_chain(mbuf_head, mbuf_tail, mbuf);
-				if (unlikely(ret < 0)) {
-					MIF_LOG(ERR, "number-of-segments-overflow");
-					rte_pktmbuf_free(mbuf);
-					goto no_free_bufs;
-				}
-			}
-			cp_len = RTE_MIN(dst_len, src_len);
-
-			rte_pktmbuf_data_len(mbuf) += cp_len;
-			rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf);
-			if (mbuf != mbuf_head)
-				rte_pktmbuf_pkt_len(mbuf_head) += cp_len;
-
-			memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off),
-			       (uint8_t *)memif_get_buffer(proc_private, d0) +
-			       src_off, cp_len);
-
-			src_off += cp_len;
-			dst_off += cp_len;
-			src_len -= cp_len;
-		} while (src_len);
-
-		cur_slot++;
-		n_slots--;
-
-		if (d0->flags & MEMIF_DESC_FLAG_NEXT)
-			goto next_slot;
-
-		mq->n_bytes += rte_pktmbuf_pkt_len(mbuf_head);
-		*bufs++ = mbuf_head;
-		n_rx_pkts++;
-	}
-
-no_free_bufs:
-	if (type == MEMIF_RING_S2M) {
-		rte_mb();
-		ring->tail = cur_slot;
-		mq->last_head = cur_slot;
-	} else {
-		mq->last_tail = cur_slot;
-	}
-
-refill:
-	if (type == MEMIF_RING_M2S) {
-		head = ring->head;
-		n_slots = ring_size - head + mq->last_tail;
-
-		while (n_slots--) {
-			s0 = head++ & mask;
-			d0 = &ring->desc[s0];
-			d0->length = pmd->run.pkt_buffer_size;
-		}
-		rte_mb();
-		ring->head = head;
-	}
-
-	mq->n_pkts += n_rx_pkts;
-	return n_rx_pkts;
-}
-
-static uint16_t
-eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
-{
-	struct memif_queue *mq = queue;
-	struct pmd_internals *pmd = rte_eth_devices[mq->in_port].data->dev_private;
-	struct pmd_process_private *proc_private =
-		rte_eth_devices[mq->in_port].process_private;
-	memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq);
-	uint16_t slot, saved_slot, n_free, ring_size, mask, n_tx_pkts = 0;
-	uint16_t src_len, src_off, dst_len, dst_off, cp_len;
-	memif_ring_type_t type = mq->type;
-	memif_desc_t *d0;
-	struct rte_mbuf *mbuf;
-	struct rte_mbuf *mbuf_head;
-	uint64_t a;
-	ssize_t size;
-	struct rte_eth_link link;
-
-	if (unlikely((pmd->flags & ETH_MEMIF_FLAG_CONNECTED) == 0))
-		return 0;
-	if (unlikely(ring == NULL)) {
-		/* Secondary process will attempt to request regions. */
-		rte_eth_link_get(mq->in_port, &link);
-		return 0;
-	}
-
-	ring_size = 1 << mq->log2_ring_size;
-	mask = ring_size - 1;
-
-	n_free = ring->tail - mq->last_tail;
-	mq->last_tail += n_free;
-	slot = (type == MEMIF_RING_S2M) ? ring->head : ring->tail;
-
-	if (type == MEMIF_RING_S2M)
-		n_free = ring_size - ring->head + mq->last_tail;
-	else
-		n_free = ring->head - ring->tail;
-
-	while (n_tx_pkts < nb_pkts && n_free) {
-		mbuf_head = *bufs++;
-		mbuf = mbuf_head;
-
-		saved_slot = slot;
-		d0 = &ring->desc[slot & mask];
-		dst_off = 0;
-		dst_len = (type == MEMIF_RING_S2M) ?
-			pmd->run.pkt_buffer_size : d0->length;
-
-next_in_chain:
-		src_off = 0;
-		src_len = rte_pktmbuf_data_len(mbuf);
-
-		while (src_len) {
-			if (dst_len == 0) {
-				if (n_free) {
-					slot++;
-					n_free--;
-					d0->flags |= MEMIF_DESC_FLAG_NEXT;
-					d0 = &ring->desc[slot & mask];
-					dst_off = 0;
-					dst_len = (type == MEMIF_RING_S2M) ?
-					    pmd->run.pkt_buffer_size : d0->length;
-					d0->flags = 0;
-				} else {
-					slot = saved_slot;
-					goto no_free_slots;
-				}
-			}
-			cp_len = RTE_MIN(dst_len, src_len);
-
-			memcpy((uint8_t *)memif_get_buffer(proc_private, d0) + dst_off,
-			       rte_pktmbuf_mtod_offset(mbuf, void *, src_off),
-			       cp_len);
-
-			mq->n_bytes += cp_len;
-			src_off += cp_len;
-			dst_off += cp_len;
-			src_len -= cp_len;
-			dst_len -= cp_len;
-
-			d0->length = dst_off;
-		}
-
-		if (rte_pktmbuf_is_contiguous(mbuf) == 0) {
-			mbuf = mbuf->next;
-			goto next_in_chain;
-		}
-
-		n_tx_pkts++;
-		slot++;
-		n_free--;
-		rte_pktmbuf_free(mbuf_head);
-	}
-
-no_free_slots:
-	rte_mb();
-	if (type == MEMIF_RING_S2M)
-		ring->head = slot;
-	else
-		ring->tail = slot;
-
-	if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0) {
-		a = 1;
-		size = write(mq->intr_handle.fd, &a, sizeof(a));
-		if (unlikely(size < 0)) {
-			MIF_LOG(WARNING,
-				"Failed to send interrupt. %s", strerror(errno));
-		}
-	}
-
-	mq->n_err += nb_pkts - n_tx_pkts;
-	mq->n_pkts += n_tx_pkts;
-	return n_tx_pkts;
-}
-
 void
-memif_free_regions(struct pmd_process_private *proc_private)
+memif_free_regions(struct rte_eth_dev *dev)
 {
+	struct pmd_process_private *proc_private = dev->process_private;
+	struct pmd_internals *pmd = dev->data->dev_private;
 	int i;
 	struct memif_region *r;

-	MIF_LOG(DEBUG, "Free memory regions");
 	/* regions are allocated contiguously, so it's
 	 * enough to loop until 'proc_private->regions_num'
 	 */
 	for (i = 0; i < proc_private->regions_num; i++) {
 		r = proc_private->regions[i];
 		if (r != NULL) {
+			/* This is memzone */
+			if (i > 0 && (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY)) {
+				r->addr = NULL;
+				if (r->fd > 0)
+					close(r->fd);
+			}
 			if (r->addr != NULL) {
 				munmap(r->addr, r->region_size);
 				if (r->fd > 0) {
@@ -511,92 +377,32 @@ memif_free_regions(struct pmd_process_private *proc_private)
 	proc_private->regions_num = 0;
 }

-static int
-memif_region_init_shm(struct rte_eth_dev *dev, uint8_t has_buffers)
-{
-	struct pmd_internals *pmd = dev->data->dev_private;
-	struct pmd_process_private *proc_private = dev->process_private;
-	char shm_name[ETH_MEMIF_SHM_NAME_SIZE];
-	int ret = 0;
-	struct memif_region *r;
-
-	if (proc_private->regions_num >= ETH_MEMIF_MAX_REGION_NUM) {
-		MIF_LOG(ERR, "Too many regions.");
-		return -1;
-	}
-
-	r = rte_zmalloc("region", sizeof(struct memif_region), 0);
-	if (r == NULL) {
-		MIF_LOG(ERR, "Failed to alloc memif region.");
-		return -ENOMEM;
-	}
-
-	/* calculate buffer offset */
-	r->pkt_buffer_offset = (pmd->run.num_s2m_rings + pmd->run.num_m2s_rings) *
-	    (sizeof(memif_ring_t) + sizeof(memif_desc_t) *
-	    (1 << pmd->run.log2_ring_size));
-
-	r->region_size = r->pkt_buffer_offset;
-	/* if region has buffers, add buffers size to region_size */
-	if (has_buffers == 1)
-		r->region_size += (uint32_t)(pmd->run.pkt_buffer_size *
-			(1 << pmd->run.log2_ring_size) *
-			(pmd->run.num_s2m_rings +
-			 pmd->run.num_m2s_rings));
-
-	memset(shm_name, 0, sizeof(char) * ETH_MEMIF_SHM_NAME_SIZE);
-	snprintf(shm_name, ETH_MEMIF_SHM_NAME_SIZE, "memif_region_%d",
-		 proc_private->regions_num);
-
-	r->fd = memfd_create(shm_name, MFD_ALLOW_SEALING);
-	if (r->fd < 0) {
-		MIF_LOG(ERR, "Failed to create shm file: %s.", strerror(errno));
-		ret = -1;
-		goto error;
-	}
-
-	ret = fcntl(r->fd, F_ADD_SEALS, F_SEAL_SHRINK);
-	if (ret < 0) {
-		MIF_LOG(ERR, "Failed to add seals to shm file: %s.", strerror(errno));
-		goto error;
-	}
-
-	ret = ftruncate(r->fd, r->region_size);
-	if (ret < 0) {
-		MIF_LOG(ERR, "Failed to truncate shm file: %s.", strerror(errno));
-		goto error;
-	}
-
-	r->addr = mmap(NULL, r->region_size, PROT_READ |
-		       PROT_WRITE, MAP_SHARED, r->fd, 0);
-	if (r->addr == MAP_FAILED) {
-		MIF_LOG(ERR, "Failed to mmap shm region: %s.", strerror(ret));
-		ret = -1;
-		goto error;
-	}
-
-	proc_private->regions[proc_private->regions_num] = r;
-	proc_private->regions_num++;
-
-	return ret;
-
-error:
-	if (r->fd > 0)
-		close(r->fd);
-	r->fd = -1;
-
-	return ret;
-}
-
 static int
 memif_regions_init(struct rte_eth_dev *dev)
 {
+	struct pmd_internals *pmd = dev->data->dev_private;
 	int ret;

-	/* create one buffer region */
-	ret = memif_region_init_shm(dev, /* has buffer */ 1);
-	if (ret < 0)
-		return ret;
+	/*
+	 * Zero-copy exposes dpdk memory.
+	 * Each memseg list will be represented by memif region.
+	 * Zero-copy regions indexing: memseg list idx + 1,
+	 * as we already have region 0 reserved for descriptors.
+	 */
+	if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY) {
+		/* create region idx 0 containing descriptors */
+		ret = memif_region_init_shm(dev, 0);
+		if (ret < 0)
+			return ret;
+		ret = rte_memseg_walk(memif_region_init_zc, (void *)dev->process_private);
+		if (ret < 0)
+			return ret;
+	} else {
+		/* create one memory region contaning rings and buffers */
+		ret = memif_region_init_shm(dev, /* has buffers */ 1);
+		if (ret < 0)
+			return ret;
+	}

 	return 0;
 }
@@ -616,6 +422,10 @@ memif_init_rings(struct rte_eth_dev *dev)
 		ring->tail = 0;
 		ring->cookie = MEMIF_COOKIE;
 		ring->flags = 0;
+
+		if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY)
+			continue;
+
 		for (j = 0; j < (1 << pmd->run.log2_ring_size); j++) {
 			slot = i * (1 << pmd->run.log2_ring_size) + j;
 			ring->desc[j].region = 0;
@@ -632,6 +442,10 @@ memif_init_rings(struct rte_eth_dev *dev)
 		ring->tail = 0;
 		ring->cookie = MEMIF_COOKIE;
 		ring->flags = 0;
+
+		if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY)
+			continue;
+
 		for (j = 0; j < (1 << pmd->run.log2_ring_size); j++) {
 			slot = (i + pmd->run.num_s2m_rings) *
 			    (1 << pmd->run.log2_ring_size) + j;
@@ -645,7 +459,7 @@ memif_init_rings(struct rte_eth_dev *dev)
 }

 /* called only by slave */
-static void
+static int
 memif_init_queues(struct rte_eth_dev *dev)
 {
 	struct pmd_internals *pmd = dev->data->dev_private;
@@ -666,6 +480,13 @@ memif_init_queues(struct rte_eth_dev *dev)
 				"Failed to create eventfd for tx queue %d: %s.", i,
 				strerror(errno));
 		}
+		mq->buffers = NULL;
+		if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY) {
+			mq->buffers = rte_zmalloc("bufs", sizeof(struct rte_mbuf *) *
+						  (1 << mq->log2_ring_size), 0);
+			if (mq->buffers == NULL)
+				return -ENOMEM;
+		}
 	}

 	for (i = 0; i < pmd->run.num_m2s_rings; i++) {
@@ -682,7 +503,15 @@ memif_init_queues(struct rte_eth_dev *dev)
 				"Failed to create eventfd for rx queue %d: %s.", i,
 				strerror(errno));
 		}
+		mq->buffers = NULL;
+		if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY) {
+			mq->buffers = rte_zmalloc("bufs", sizeof(struct rte_mbuf *) *
+						  (1 << mq->log2_ring_size), 0);
+			if (mq->buffers == NULL)
+				return -ENOMEM;
+		}
 	}
+	return 0;
 }

 int
@@ -696,7 +525,9 @@ memif_init_regions_and_queues(struct rte_eth_dev *dev)

 	memif_init_rings(dev);

-	memif_init_queues(dev);
+	ret = memif_init_queues(dev);
+	if (ret < 0)
+		return ret;

 	return 0;
 }
@@ -720,8 +551,16 @@ memif_connect(struct rte_eth_dev *dev)
 				mr->addr = mmap(NULL, mr->region_size,
 						PROT_READ | PROT_WRITE,
 						MAP_SHARED, mr->fd, 0);
-				if (mr->addr == NULL)
+				if (mr->addr == MAP_FAILED) {
+					MIF_LOG(ERR, "mmap failed: %s\n",
+						strerror(errno));
 					return -1;
+				}
+			}
+			if (i > 0 && (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY)) {
+				/* close memseg file */
+				close(mr->fd);
+				mr->fd = -1;
 			}
 		}
 	}
@@ -782,8 +621,7 @@ memif_dev_start(struct rte_eth_dev *dev)
 		ret = memif_connect_master(dev);
 		break;
 	default:
-		MIF_LOG(ERR, "%s: Unknown role: %d.",
-			rte_vdev_device_name(pmd->vdev), pmd->role);
+		MIF_LOG(ERR, "Unknown role: %d.", pmd->role);
 		ret = -1;
 		break;
 	}
@@ -848,8 +686,7 @@ memif_tx_queue_setup(struct rte_eth_dev *dev,

 	mq = rte_zmalloc("tx-queue", sizeof(struct memif_queue), 0);
 	if (mq == NULL) {
-		MIF_LOG(ERR, "%s: Failed to allocate tx queue id: %u",
-			rte_vdev_device_name(pmd->vdev), qid);
+		MIF_LOG(ERR, "Failed to allocate tx queue id: %u", qid);
 		return -ENOMEM;
 	}

@@ -878,8 +715,7 @@ memif_rx_queue_setup(struct rte_eth_dev *dev,

 	mq = rte_zmalloc("rx-queue", sizeof(struct memif_queue), 0);
 	if (mq == NULL) {
-		MIF_LOG(ERR, "%s: Failed to allocate rx queue id: %u",
-			rte_vdev_device_name(pmd->vdev), qid);
+		MIF_LOG(ERR, "Failed to allocate rx queue id: %u", qid);
 		return -ENOMEM;
 	}

@@ -920,7 +756,7 @@ memif_link_update(struct rte_eth_dev *dev,
 			memif_mp_request_regions(dev);
 		} else if (dev->data->dev_link.link_status == ETH_LINK_DOWN &&
 				proc_private->regions_num > 0) {
-			memif_free_regions(proc_private);
+			memif_free_regions(dev);
 		}
 	}
 	return 0;
@@ -964,6 +800,7 @@ memif_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 		mq = dev->data->tx_queues[i];
 		stats->q_opackets[i] = mq->n_pkts;
 		stats->q_obytes[i] = mq->n_bytes;
+		stats->q_errors[i] = mq->n_err;
 		stats->opackets += mq->n_pkts;
 		stats->obytes += mq->n_bytes;
 		stats->oerrors += mq->n_err;
@@ -1043,11 +880,6 @@ memif_create(struct rte_vdev_device *vdev, enum memif_role_t role,
 	const unsigned int numa_node = vdev->device.numa_node;
 	const char *name = rte_vdev_device_name(vdev);

-	if (flags & ETH_MEMIF_FLAG_ZERO_COPY) {
-		MIF_LOG(ERR, "Zero-copy slave not supported.");
-		return -1;
-	}
-
 	eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
 	if (eth_dev == NULL) {
 		MIF_LOG(ERR, "%s: Unable to allocate device struct.", name);
@@ -1071,6 +903,9 @@ memif_create(struct rte_vdev_device *vdev, enum memif_role_t role,
 	pmd->flags = flags;
 	pmd->flags |= ETH_MEMIF_FLAG_DISABLED;
 	pmd->role = role;
+	/* Zero-copy flag irelevant to master. */
+	if (pmd->role == MEMIF_ROLE_MASTER)
+		pmd->flags &= ~ETH_MEMIF_FLAG_ZERO_COPY;

 	ret = memif_socket_init(eth_dev, socket_filename);
 	if (ret < 0)
@@ -1094,8 +929,14 @@ memif_create(struct rte_vdev_device *vdev, enum memif_role_t role,

 	eth_dev->dev_ops = &ops;
 	eth_dev->device = &vdev->device;
-	eth_dev->rx_pkt_burst = eth_memif_rx;
-	eth_dev->tx_pkt_burst = eth_memif_tx;
+	if (pmd->flags & ETH_MEMIF_FLAG_ZERO_COPY) {
+		eth_dev->rx_pkt_burst = eth_memif_rx_zc;
+		eth_dev->tx_pkt_burst = eth_memif_tx_zc;
+	} else {
+		eth_dev->rx_pkt_burst = eth_memif_rx;
+		eth_dev->tx_pkt_burst = eth_memif_tx;
+	}
+

 	eth_dev->data->dev_flags &= RTE_ETH_DEV_CLOSE_REMOVE;

@@ -1124,9 +965,14 @@ memif_set_role(const char *key __rte_unused, const char *value,
 static int
 memif_set_zc(const char *key __rte_unused, const char *value, void *extra_args)
 {
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
 	uint32_t *flags = (uint32_t *)extra_args;

 	if (strstr(value, "yes") != NULL) {
+		if (!mcfg->single_file_segments) {
+			MIF_LOG(ERR, "Zero-copy doesn't support multi-file segments.");
+			return -ENOTSUP;
+		}
 		*flags |= ETH_MEMIF_FLAG_ZERO_COPY;
 	} else if (strstr(value, "no") != NULL) {
 		*flags &= ~ETH_MEMIF_FLAG_ZERO_COPY;
diff --git a/drivers/net/memif/rte_eth_memif.h b/drivers/net/memif/rte_eth_memif.h
index 24e8a0914..dea79820c 100644
--- a/drivers/net/memif/rte_eth_memif.h
+++ b/drivers/net/memif/rte_eth_memif.h
@@ -63,13 +63,16 @@ struct memif_queue {
 	uint16_t last_head;			/**< last ring head */
 	uint16_t last_tail;			/**< last ring tail */

+	struct rte_mbuf **buffers;
+	/**< Stored mbufs. Used in zero-copy tx. Slave stores transmitted
+	 * mbufs to free them once master has received them.
+	 */
+
 	/* rx/tx info */
 	uint64_t n_pkts;			/**< number of rx/tx packets */
 	uint64_t n_bytes;			/**< number of rx/tx bytes */
 	uint64_t n_err;				/**< number of tx errors */

-	memif_ring_t *ring;			/**< pointer to ring */
-
 	struct rte_intr_handle intr_handle;	/**< interrupt handle */

 	memif_log2_ring_size_t log2_ring_size;	/**< log2 of ring size */
@@ -132,7 +135,7 @@ struct pmd_process_private {
  * @param proc_private
  *   device process private data
  */
-void memif_free_regions(struct pmd_process_private *proc_private);
+void memif_free_regions(struct rte_eth_dev *dev);

 /**
  * Finalize connection establishment process. Map shared memory file
@@ -158,6 +161,9 @@ int memif_connect(struct rte_eth_dev *dev);
  */
 int memif_init_regions_and_queues(struct rte_eth_dev *dev);

+memif_ring_t *memif_get_ring_from_queue(struct pmd_process_private *proc_private,
+			  struct memif_queue *mq);
+
 /**
  * Get memif version string.
  *
--
2.17.1


More information about the dev mailing list