From: Yaacov Hazan <yaac...@mellanox.com>
VLAN insertion is done in software by the PMD by default unless
CONFIG_RTE_LIBRTE_MLX5_VERBS_VLAN_INSERTION is enabled and Verbs provides
support for hardware insertion.
When enabled, this option improves performance when VLAN insertion is
requested, however ConnectX-4 Lx boards cannot take advantage of
multi-packet send optimizations anymore.
Signed-off-by: Yaacov Hazan <yaacovh at mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil at 6wind.com>
---
config/common_linuxapp | 1 +
doc/guides/nics/mlx5.rst | 8 +++
doc/guides/rel_notes/release_16_04.rst | 4 ++
drivers/net/mlx5/Makefile | 9 +++
drivers/net/mlx5/mlx5_defs.h | 9 +++
drivers/net/mlx5/mlx5_ethdev.c | 12 ++--
drivers/net/mlx5/mlx5_rxtx.c | 109 +++++++++++++++++++++++++++------
drivers/net/mlx5/mlx5_rxtx.h | 13 ++++
drivers/net/mlx5/mlx5_txq.c | 15 ++++-
9 files changed, 155 insertions(+), 25 deletions(-)
diff --git a/config/common_linuxapp b/config/common_linuxapp
index 7b5e49f..793d262 100644
--- a/config/common_linuxapp
+++ b/config/common_linuxapp
@@ -220,6 +220,7 @@ CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N=4
CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE=0
CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE=8
+CONFIG_RTE_LIBRTE_MLX5_VERBS_VLAN_INSERTION=n
#
# Compile burst-oriented Broadcom PMD driver
diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 66fe0d9..1ca1534 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -84,6 +84,7 @@ Features
- Support for multiple MAC addresses.
- VLAN filtering.
- RX VLAN stripping.
+- TX VLAN insertion.
- Promiscuous mode.
- Multicast promiscuous mode.
- Hardware checksum offloads.
@@ -142,6 +143,13 @@ These options can be modified in the ``.config`` file.
This value is always 1 for RX queues since they use a single MP.
+- ``CONFIG_RTE_LIBRTE_MLX5_VERBS_VLAN_INSERTION`` (default **n**)
+
+ Use Verbs instead of PMD implementation for VLAN insertion. Disabled by
+ default since it prevents ConnectX-4 Lx adapters from taking advantage of
+ multi-packet send optimizations, otherwise provides better performance
+ when VLAN insertion is requested.
+
Environment variables
~~~~~~~~~~~~~~~~~~~~~
diff --git a/doc/guides/rel_notes/release_16_04.rst
b/doc/guides/rel_notes/release_16_04.rst
index a3a30fd..906d835 100644
--- a/doc/guides/rel_notes/release_16_04.rst
+++ b/doc/guides/rel_notes/release_16_04.rst
@@ -92,6 +92,10 @@ This section should contain new features added in this
release. Sample format:
Added an option to make PCI bus transactions rounded to multiple of 64
bytes for better cache alignment.
+* **mlx5: TX VLAN insertion support.**
+
+ Added support for TX VLAN insertion.
+
Resolved Issues
---------------
diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 712c0a9..a260c52 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -101,6 +101,10 @@ ifdef CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE
CFLAGS += -DMLX5_PMD_TX_MP_CACHE=$(CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE)
endif
+ifeq ($(CONFIG_RTE_LIBRTE_MLX5_VERBS_VLAN_INSERTION),y)
+CFLAGS += -DMLX5_VERBS_VLAN_INSERTION
+endif
+
include $(RTE_SDK)/mk/rte.lib.mk
# Generate and clean-up mlx5_autoconf.h.
@@ -142,6 +146,11 @@ mlx5_autoconf.h: $(RTE_SDK)/scripts/auto-config-h.sh
infiniband/verbs.h \
enum IBV_EXP_CREATE_WQ_FLAG_RX_END_PADDING \
$(AUTOCONF_OUTPUT)
+ $Q sh -- '$<' '$@' \
+ HAVE_VERBS_VLAN_INSERTION \
+ infiniband/verbs.h \
+ enum IBV_EXP_RECEIVE_WQ_CVLAN_INSERTION \
+ $(AUTOCONF_OUTPUT)
$(SRCS-$(CONFIG_RTE_LIBRTE_MLX5_PMD):.c=.o): mlx5_autoconf.h
diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index 5b00d8e..fb8db2e 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -95,4 +95,13 @@
#define MLX5_FDIR_SUPPORT 1
#endif
+/*
+ * Prevent compilation when HW VLAN insertion is requested by configuration
+ * but not supported by Verbs.
+ */
+#if defined(MLX5_VERBS_VLAN_INSERTION) && !defined(HAVE_VERBS_VLAN_INSERTION)
+#error CONFIG_RTE_LIBRTE_MLX5_VERBS_VLAN_INSERTION \
+ enabled in configuration but not supported by libibverbs.
+#endif
+
#endif /* RTE_PMD_MLX5_DEFS_H_ */
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 6b674a2..66115d2 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -544,12 +544,12 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct
rte_eth_dev_info *info)
DEV_RX_OFFLOAD_UDP_CKSUM |
DEV_RX_OFFLOAD_TCP_CKSUM) :
0);
- info->tx_offload_capa =
- (priv->hw_csum ?
- (DEV_TX_OFFLOAD_IPV4_CKSUM |
- DEV_TX_OFFLOAD_UDP_CKSUM |
- DEV_TX_OFFLOAD_TCP_CKSUM) :
- 0);
+ info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
+ if (priv->hw_csum)
+ info->tx_offload_capa |=
+ (DEV_TX_OFFLOAD_IPV4_CKSUM |
+ DEV_TX_OFFLOAD_UDP_CKSUM |
+ DEV_TX_OFFLOAD_TCP_CKSUM);
if (priv_get_ifname(priv, &ifname) == 0)
info->if_index = if_nametoindex(ifname);
/* FIXME: RETA update/query API expects the callee to know the size of
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 4919189..9fc535e 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -333,6 +333,40 @@ txq_mp2mr_iter(const struct rte_mempool *mp, void *arg)
txq_mp2mr(txq, mp);
}
+#ifndef MLX5_VERBS_VLAN_INSERTION
+
+/**
+ * Insert VLAN to specific packet, using the mbuf's headroom space.
+ *
+ * @param buf
+ * Buffer to insert the vlan.
+ *
+ * @return
+ * 0 on success, errno value on failure.
+ */
+static inline int
+insert_vlan_sw(struct rte_mbuf *buf)
+{
+ uintptr_t addr;
+ uint32_t vlan;
+ uint16_t head_room_len = rte_pktmbuf_headroom(buf);
+
+ if (head_room_len < 4)
+ return EINVAL;
+
+ addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ vlan = htonl(0x81000000 | buf->vlan_tci);
+ memmove((void *)(addr - 4), (void *)addr, 12);
+ memcpy((void *)(addr + 8), &vlan, sizeof(vlan));
+
+ SET_DATA_OFF(buf, head_room_len - 4);
+ DATA_LEN(buf) += 4;
+
+ return 0;
+}
+
+#endif /* !MLX5_VERBS_VLAN_INSERTION */
+
#if MLX5_PMD_SGE_WR_N > 1
/**
@@ -554,6 +588,14 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts,
uint16_t pkts_n)
if (RTE_ETH_IS_TUNNEL_PKT(buf->packet_type))
send_flags |= IBV_EXP_QP_BURST_TUNNEL;
}
+#ifndef MLX5_VERBS_VLAN_INSERTION
+ if (buf->ol_flags & PKT_TX_VLAN_PKT) {
+ err = insert_vlan_sw(buf);
+
+ if (unlikely(err))
+ goto stop;
+ }
+#endif /* !MLX5_VERBS_VLAN_INSERTION */
if (likely(segs == 1)) {
uintptr_t addr;
uint32_t length;
@@ -577,13 +619,23 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts,
uint16_t pkts_n)
}
/* Put packet into send queue. */
#if MLX5_PMD_MAX_INLINE > 0
- if (length <= txq->max_inline)
- err = txq->send_pending_inline
- (txq->qp,
- (void *)addr,
- length,
- send_flags);
- else
+ if (length <= txq->max_inline) {
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ if (buf->ol_flags & PKT_TX_VLAN_PKT)
+ err = txq->send_pending_inline_vlan
+ (txq->qp,
+ (void *)addr,
+ length,
+ send_flags,
+ &buf->vlan_tci);
+ else
+#endif /* MLX5_VERBS_VLAN_INSERTION */
+ err = txq->send_pending_inline
+ (txq->qp,
+ (void *)addr,
+ length,
+ send_flags);
+ } else
#endif
{
/* Retrieve Memory Region key for this
@@ -597,12 +649,23 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts,
uint16_t pkts_n)
elt->buf = NULL;
goto stop;
}
- err = txq->send_pending
- (txq->qp,
- addr,
- length,
- lkey,
- send_flags);
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ if (buf->ol_flags & PKT_TX_VLAN_PKT)
+ err = txq->send_pending_vlan
+ (txq->qp,
+ addr,
+ length,
+ lkey,
+ send_flags,
+ &buf->vlan_tci);
+ else
+#endif /* MLX5_VERBS_VLAN_INSERTION */
+ err = txq->send_pending
+ (txq->qp,
+ addr,
+ length,
+ lkey,
+ send_flags);
}
if (unlikely(err))
goto stop;
@@ -619,11 +682,21 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts,
uint16_t pkts_n)
if (ret.length == (unsigned int)-1)
goto stop;
/* Put SG list into send queue. */
- err = txq->send_pending_sg_list
- (txq->qp,
- sges,
- ret.num,
- send_flags);
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ if (buf->ol_flags & PKT_TX_VLAN_PKT)
+ err = txq->send_pending_sg_list_vlan
+ (txq->qp,
+ sges,
+ ret.num,
+ send_flags,
+ &buf->vlan_tci);
+ else
+#endif /* MLX5_VERBS_VLAN_INSERTION */
+ err = txq->send_pending_sg_list
+ (txq->qp,
+ sges,
+ ret.num,
+ send_flags);
if (unlikely(err))
goto stop;
#ifdef MLX5_PMD_SOFT_COUNTERS
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 6a0087e..7306d18 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -254,11 +254,20 @@ struct txq {
struct priv *priv; /* Back pointer to private data. */
int32_t (*poll_cnt)(struct ibv_cq *cq, uint32_t max);
int (*send_pending)();
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ int (*send_pending_vlan)();
+#endif
#if MLX5_PMD_MAX_INLINE > 0
int (*send_pending_inline)();
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ int (*send_pending_inline_vlan)();
+#endif
#endif
#if MLX5_PMD_SGE_WR_N > 1
int (*send_pending_sg_list)();
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ int (*send_pending_sg_list_vlan)();
+#endif
#endif
int (*send_flush)(struct ibv_qp *qp);
struct ibv_cq *cq; /* Completion Queue. */
@@ -282,7 +291,11 @@ struct txq {
/* Elements used only for init part are here. */
linear_t (*elts_linear)[]; /* Linearized buffers. */
struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
+#ifdef HAVE_VERBS_VLAN_INSERTION
+ struct ibv_exp_qp_burst_family_v1 *if_qp; /* QP burst interface. */
+#else
struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
+#endif
struct ibv_exp_cq_family *if_cq; /* CQ interface. */
struct ibv_exp_res_domain *rd; /* Resource Domain. */
unsigned int socket; /* CPU socket ID for allocations. */
diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
index 6700af4..c643cf4 100644
--- a/drivers/net/mlx5/mlx5_txq.c
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -400,7 +400,11 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq,
uint16_t desc,
.intf_scope = IBV_EXP_INTF_GLOBAL,
.intf = IBV_EXP_INTF_QP_BURST,
.obj = tmpl.qp,
-#ifdef HAVE_EXP_QP_BURST_CREATE_ENABLE_MULTI_PACKET_SEND_WR
+#ifdef HAVE_VERBS_VLAN_INSERTION
+ .intf_version = 1,
+#endif
+#if defined(HAVE_EXP_QP_BURST_CREATE_ENABLE_MULTI_PACKET_SEND_WR) && \
+ !defined(MLX5_VERBS_VLAN_INSERTION)
/* Multi packet send WR can only be used outside of VF. */
.family_flags =
(!priv->vf ?
@@ -422,11 +426,20 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq,
uint16_t desc,
txq->poll_cnt = txq->if_cq->poll_cnt;
#if MLX5_PMD_MAX_INLINE > 0
txq->send_pending_inline = txq->if_qp->send_pending_inline;
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ txq->send_pending_inline_vlan = txq->if_qp->send_pending_inline_vlan;
+#endif
#endif
#if MLX5_PMD_SGE_WR_N > 1
txq->send_pending_sg_list = txq->if_qp->send_pending_sg_list;
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ txq->send_pending_sg_list_vlan = txq->if_qp->send_pending_sg_list_vlan;
+#endif
#endif
txq->send_pending = txq->if_qp->send_pending;
+#ifdef MLX5_VERBS_VLAN_INSERTION
+ txq->send_pending_vlan = txq->if_qp->send_pending_vlan;
+#endif
txq->send_flush = txq->if_qp->send_flush;
DEBUG("%p: txq updated with %p", (void *)txq, (void *)&tmpl);
/* Pre-register known mempools. */
--
2.1.4
