From dbccb4cddcd2f7cff7d6ddcf11f5aeedc6245ab3 Mon Sep 17 00:00:00 2001 From: Shahaf Shuler Date: Wed, 10 Jan 2018 11:17:00 +0200 Subject: [PATCH] net/mlx5: convert to new Tx offloads API Ethdev Tx offloads API has changed since: commit cba7f53b717d ("ethdev: introduce Tx queue offloads API") This commit support the new Tx offloads API. Signed-off-by: Shahaf Shuler Acked-by: Nelio Laranjeiro --- doc/guides/nics/mlx5.rst | 15 +++--- drivers/net/mlx5/mlx5.c | 18 ++----- drivers/net/mlx5/mlx5.h | 2 +- drivers/net/mlx5/mlx5_ethdev.c | 37 ++++++++------- drivers/net/mlx5/mlx5_rxtx.c | 6 ++- drivers/net/mlx5/mlx5_rxtx.h | 7 +-- drivers/net/mlx5/mlx5_rxtx_vec.c | 32 ++++++------- drivers/net/mlx5/mlx5_rxtx_vec.h | 12 +++++ drivers/net/mlx5/mlx5_txq.c | 80 ++++++++++++++++++++++++++++++-- 9 files changed, 142 insertions(+), 67 deletions(-) diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst index 154db64d7b..bdc2216c03 100644 --- a/doc/guides/nics/mlx5.rst +++ b/doc/guides/nics/mlx5.rst @@ -262,8 +262,9 @@ Run-time configuration Enhanced MPS supports hybrid mode - mixing inlined packets and pointers in the same descriptor. - This option cannot be used in conjunction with ``tso`` below. When ``tso`` - is set, ``txq_mpw_en`` is disabled. + This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO, + DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``. + When those offloads are requested the MPS send function will not be used. It is currently only supported on the ConnectX-4 Lx and ConnectX-5 families of adapters. Enabled by default. @@ -284,17 +285,15 @@ Run-time configuration Effective only when Enhanced MPS is supported. The default value is 256. -- ``tso`` parameter [int] - - A nonzero value enables hardware TSO. - When hardware TSO is enabled, packets marked with TCP segmentation - offload will be divided into segments by the hardware. Disabled by default. - - ``tx_vec_en`` parameter [int] A nonzero value enables Tx vector on ConnectX-5 only NIC if the number of global Tx queues on the port is lesser than MLX5_VPMD_MIN_TXQS. + This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO, + DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``. + When those offloads are requested the MPS send function will not be used. + Enabled by default on ConnectX-5. - ``rx_vec_en`` parameter [int] diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c index ca44a0a593..1c95f3520e 100644 --- a/drivers/net/mlx5/mlx5.c +++ b/drivers/net/mlx5/mlx5.c @@ -85,9 +85,6 @@ /* Device parameter to limit the size of inlining packet. */ #define MLX5_TXQ_MAX_INLINE_LEN "txq_max_inline_len" -/* Device parameter to enable hardware TSO offload. */ -#define MLX5_TSO "tso" - /* Device parameter to enable hardware Tx vector. */ #define MLX5_TX_VEC_EN "tx_vec_en" @@ -406,8 +403,6 @@ mlx5_args_check(const char *key, const char *val, void *opaque) config->mpw_hdr_dseg = !!tmp; } else if (strcmp(MLX5_TXQ_MAX_INLINE_LEN, key) == 0) { config->inline_max_packet_sz = tmp; - } else if (strcmp(MLX5_TSO, key) == 0) { - config->tso = !!tmp; } else if (strcmp(MLX5_TX_VEC_EN, key) == 0) { config->tx_vec_en = !!tmp; } else if (strcmp(MLX5_RX_VEC_EN, key) == 0) { @@ -440,7 +435,6 @@ mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs) MLX5_TXQ_MPW_EN, MLX5_TXQ_MPW_HDR_DSEG_EN, MLX5_TXQ_MAX_INLINE_LEN, - MLX5_TSO, MLX5_TX_VEC_EN, MLX5_RX_VEC_EN, NULL, @@ -629,7 +623,6 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) .cqe_comp = cqe_comp, .mps = mps, .tunnel_en = tunnel_en, - .tso = 0, .tx_vec_en = 1, .rx_vec_en = 1, .mpw_hdr_dseg = 0, @@ -793,10 +786,9 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) priv_get_num_vfs(priv, &num_vfs); config.sriov = (num_vfs || sriov); - if (config.tso) - config.tso = ((device_attr_ex.tso_caps.max_tso > 0) && - (device_attr_ex.tso_caps.supported_qpts & - (1 << IBV_QPT_RAW_PACKET))); + config.tso = ((device_attr_ex.tso_caps.max_tso > 0) && + (device_attr_ex.tso_caps.supported_qpts & + (1 << IBV_QPT_RAW_PACKET))); if (config.tso) config.tso_max_payload_sz = device_attr_ex.tso_caps.max_tso; @@ -805,10 +797,6 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) " (" MLX5_TXQ_MPW_EN ")"); err = ENOTSUP; goto port_error; - } else if (config.mps && config.tso) { - WARN("multi-packet send not supported in conjunction " - "with TSO. MPS disabled"); - config.mps = 0; } INFO("%sMPS is %s", config.mps == MLX5_MPW_ENHANCED ? "Enhanced " : "", diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h index 171b3a9338..8ee5220698 100644 --- a/drivers/net/mlx5/mlx5.h +++ b/drivers/net/mlx5/mlx5.h @@ -112,7 +112,7 @@ struct mlx5_dev_config { unsigned int tunnel_en:1; /* Whether tunnel is supported. */ unsigned int flow_counter_en:1; /* Whether flow counter is supported. */ unsigned int cqe_comp:1; /* CQE compression is enabled. */ - unsigned int tso:1; /* Whether TSO is enabled. */ + unsigned int tso:1; /* Whether TSO is supported. */ unsigned int tx_vec_en:1; /* Tx vector is enabled. */ unsigned int rx_vec_en:1; /* Rx vector is enabled. */ unsigned int mpw_hdr_dseg:1; /* Enable DSEGs in the title WQEBB. */ diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c index d2f98769e7..7b1b7aa0e6 100644 --- a/drivers/net/mlx5/mlx5_ethdev.c +++ b/drivers/net/mlx5/mlx5_ethdev.c @@ -551,7 +551,15 @@ dev_configure(struct rte_eth_dev *dev) unsigned int reta_idx_n; const uint8_t use_app_rss_key = !!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key; - + uint64_t supp_tx_offloads = mlx5_priv_get_tx_port_offloads(priv); + uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads; + + if ((tx_offloads & supp_tx_offloads) != tx_offloads) { + ERROR("Some Tx offloads are not supported " + "requested 0x%" PRIx64 " supported 0x%" PRIx64, + tx_offloads, supp_tx_offloads); + return ENOTSUP; + } if (use_app_rss_key && (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len != rss_hash_default_key_len)) { @@ -672,19 +680,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) (priv->config.hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0) | DEV_RX_OFFLOAD_TIMESTAMP; - if (!config->mps) - info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT; - if (config->hw_csum) - info->tx_offload_capa |= - (DEV_TX_OFFLOAD_IPV4_CKSUM | - DEV_TX_OFFLOAD_UDP_CKSUM | - DEV_TX_OFFLOAD_TCP_CKSUM); - if (config->tso) - info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO; - if (config->tunnel_en) - info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM | - DEV_TX_OFFLOAD_VXLAN_TNL_TSO | - DEV_TX_OFFLOAD_GRE_TNL_TSO); + info->tx_offload_capa = mlx5_priv_get_tx_port_offloads(priv); if (priv_get_ifname(priv, &ifname) == 0) info->if_index = if_nametoindex(ifname); info->reta_size = priv->reta_idx_n ? @@ -1392,16 +1388,23 @@ mlx5_set_link_up(struct rte_eth_dev *dev) * Pointer to selected Tx burst function. */ eth_tx_burst_t -priv_select_tx_function(struct priv *priv, __rte_unused struct rte_eth_dev *dev) +priv_select_tx_function(struct priv *priv, struct rte_eth_dev *dev) { eth_tx_burst_t tx_pkt_burst = mlx5_tx_burst; struct mlx5_dev_config *config = &priv->config; + uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads; + int tso = !!(tx_offloads & (DEV_TX_OFFLOAD_TCP_TSO | + DEV_TX_OFFLOAD_VXLAN_TNL_TSO | + DEV_TX_OFFLOAD_GRE_TNL_TSO)); + int vlan_insert = !!(tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT); assert(priv != NULL); /* Select appropriate TX function. */ + if (vlan_insert || tso) + return tx_pkt_burst; if (config->mps == MLX5_MPW_ENHANCED) { - if (priv_check_vec_tx_support(priv) > 0) { - if (priv_check_raw_vec_tx_support(priv) > 0) + if (priv_check_vec_tx_support(priv, dev) > 0) { + if (priv_check_raw_vec_tx_support(priv, dev) > 0) tx_pkt_burst = mlx5_tx_burst_raw_vec; else tx_pkt_burst = mlx5_tx_burst_vec; diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c index 67e3db168b..3b8f71c281 100644 --- a/drivers/net/mlx5/mlx5_rxtx.c +++ b/drivers/net/mlx5/mlx5_rxtx.c @@ -1994,16 +1994,18 @@ mlx5_rx_burst_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) } int __attribute__((weak)) -priv_check_raw_vec_tx_support(struct priv *priv) +priv_check_raw_vec_tx_support(struct priv *priv, struct rte_eth_dev *dev) { (void)priv; + (void)dev; return -ENOTSUP; } int __attribute__((weak)) -priv_check_vec_tx_support(struct priv *priv) +priv_check_vec_tx_support(struct priv *priv, struct rte_eth_dev *dev) { (void)priv; + (void)dev; return -ENOTSUP; } diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h index eb7a3db85c..f7ff554b0d 100644 --- a/drivers/net/mlx5/mlx5_rxtx.h +++ b/drivers/net/mlx5/mlx5_rxtx.h @@ -200,7 +200,7 @@ struct mlx5_txq_data { uint16_t inline_max_packet_sz; /* Max packet size for inlining. */ uint16_t mr_cache_idx; /* Index of last hit entry. */ uint32_t qp_num_8s; /* QP number shifted by 8. */ - uint32_t flags; /* Flags for Tx Queue. */ + uint64_t offloads; /* Offloads for Tx Queue. */ volatile struct mlx5_cqe (*cqes)[]; /* Completion queue. */ volatile void *wqes; /* Work queue (use volatile to write into). */ volatile uint32_t *qp_db; /* Work queue doorbell. */ @@ -292,6 +292,7 @@ int mlx5_priv_txq_release(struct priv *, uint16_t); int mlx5_priv_txq_releasable(struct priv *, uint16_t); int mlx5_priv_txq_verify(struct priv *); void txq_alloc_elts(struct mlx5_txq_ctrl *); +uint64_t mlx5_priv_get_tx_port_offloads(struct priv *); /* mlx5_rxtx.c */ @@ -309,8 +310,8 @@ int mlx5_rx_descriptor_status(void *, uint16_t); int mlx5_tx_descriptor_status(void *, uint16_t); /* Vectorized version of mlx5_rxtx.c */ -int priv_check_raw_vec_tx_support(struct priv *); -int priv_check_vec_tx_support(struct priv *); +int priv_check_raw_vec_tx_support(struct priv *, struct rte_eth_dev *); +int priv_check_vec_tx_support(struct priv *, struct rte_eth_dev *); int rxq_check_vec_support(struct mlx5_rxq_data *); int priv_check_vec_rx_support(struct priv *); uint16_t mlx5_tx_burst_raw_vec(void *, struct rte_mbuf **, uint16_t); diff --git a/drivers/net/mlx5/mlx5_rxtx_vec.c b/drivers/net/mlx5/mlx5_rxtx_vec.c index ae398dcb2a..142c786d5b 100644 --- a/drivers/net/mlx5/mlx5_rxtx_vec.c +++ b/drivers/net/mlx5/mlx5_rxtx_vec.c @@ -160,15 +160,15 @@ mlx5_tx_burst_vec(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) uint16_t ret; /* Transmit multi-seg packets in the head of pkts list. */ - if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS) && + if ((txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS) && NB_SEGS(pkts[nb_tx]) > 1) nb_tx += txq_scatter_v(txq, &pkts[nb_tx], pkts_n - nb_tx); n = RTE_MIN((uint16_t)(pkts_n - nb_tx), MLX5_VPMD_TX_MAX_BURST); - if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS)) + if (txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS) n = txq_count_contig_single_seg(&pkts[nb_tx], n); - if (!(txq->flags & ETH_TXQ_FLAGS_NOOFFLOADS)) + if (txq->offloads & MLX5_VEC_TX_CKSUM_OFFLOAD_CAP) n = txq_calc_offload(txq, &pkts[nb_tx], n, &cs_flags); ret = txq_burst_v(txq, &pkts[nb_tx], n, cs_flags); nb_tx += ret; @@ -253,24 +253,20 @@ mlx5_rx_burst_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) * * @param priv * Pointer to private structure. + * @param dev + * Pointer to rte_eth_dev structure. * * @return * 1 if supported, negative errno value if not. */ int __attribute__((cold)) -priv_check_raw_vec_tx_support(struct priv *priv) +priv_check_raw_vec_tx_support(__rte_unused struct priv *priv, + struct rte_eth_dev *dev) { - uint16_t i; - - /* All the configured queues should support. */ - for (i = 0; i < priv->txqs_n; ++i) { - struct mlx5_txq_data *txq = (*priv->txqs)[i]; + uint64_t offloads = dev->data->dev_conf.txmode.offloads; - if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS) || - !(txq->flags & ETH_TXQ_FLAGS_NOOFFLOADS)) - break; - } - if (i != priv->txqs_n) + /* Doesn't support any offload. */ + if (offloads) return -ENOTSUP; return 1; } @@ -280,17 +276,21 @@ priv_check_raw_vec_tx_support(struct priv *priv) * * @param priv * Pointer to private structure. + * @param dev + * Pointer to rte_eth_dev structure. * * @return * 1 if supported, negative errno value if not. */ int __attribute__((cold)) -priv_check_vec_tx_support(struct priv *priv) +priv_check_vec_tx_support(struct priv *priv, struct rte_eth_dev *dev) { + uint64_t offloads = dev->data->dev_conf.txmode.offloads; + if (!priv->config.tx_vec_en || priv->txqs_n > MLX5_VPMD_MIN_TXQS || priv->config.mps != MLX5_MPW_ENHANCED || - priv->config.tso) + offloads & ~MLX5_VEC_TX_OFFLOAD_CAP) return -ENOTSUP; return 1; } diff --git a/drivers/net/mlx5/mlx5_rxtx_vec.h b/drivers/net/mlx5/mlx5_rxtx_vec.h index 1f08ed0b27..7d7f016f16 100644 --- a/drivers/net/mlx5/mlx5_rxtx_vec.h +++ b/drivers/net/mlx5/mlx5_rxtx_vec.h @@ -40,6 +40,18 @@ #include "mlx5_autoconf.h" #include "mlx5_prm.h" +/* HW checksum offload capabilities of vectorized Tx. */ +#define MLX5_VEC_TX_CKSUM_OFFLOAD_CAP \ + (DEV_TX_OFFLOAD_IPV4_CKSUM | \ + DEV_TX_OFFLOAD_UDP_CKSUM | \ + DEV_TX_OFFLOAD_TCP_CKSUM | \ + DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) + +/* HW offload capabilities of vectorized Tx. */ +#define MLX5_VEC_TX_OFFLOAD_CAP \ + (MLX5_VEC_TX_CKSUM_OFFLOAD_CAP | \ + DEV_TX_OFFLOAD_MULTI_SEGS) + /* * Compile time sanity check for vectorized functions. */ diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c index 3e2075c795..26db15a4f2 100644 --- a/drivers/net/mlx5/mlx5_txq.c +++ b/drivers/net/mlx5/mlx5_txq.c @@ -115,6 +115,63 @@ txq_free_elts(struct mlx5_txq_ctrl *txq_ctrl) } } +/** + * Returns the per-port supported offloads. + * + * @param priv + * Pointer to private structure. + * + * @return + * Supported Tx offloads. + */ +uint64_t +mlx5_priv_get_tx_port_offloads(struct priv *priv) +{ + uint64_t offloads = (DEV_TX_OFFLOAD_MULTI_SEGS | + DEV_TX_OFFLOAD_VLAN_INSERT); + struct mlx5_dev_config *config = &priv->config; + + if (config->hw_csum) + offloads |= (DEV_TX_OFFLOAD_IPV4_CKSUM | + DEV_TX_OFFLOAD_UDP_CKSUM | + DEV_TX_OFFLOAD_TCP_CKSUM); + if (config->tso) + offloads |= DEV_TX_OFFLOAD_TCP_TSO; + if (config->tunnel_en) { + if (config->hw_csum) + offloads |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM; + if (config->tso) + offloads |= (DEV_TX_OFFLOAD_VXLAN_TNL_TSO | + DEV_TX_OFFLOAD_GRE_TNL_TSO); + } + return offloads; +} + +/** + * Checks if the per-queue offload configuration is valid. + * + * @param priv + * Pointer to private structure. + * @param offloads + * Per-queue offloads configuration. + * + * @return + * 1 if the configuration is valid, 0 otherwise. + */ +static int +priv_is_tx_queue_offloads_allowed(struct priv *priv, uint64_t offloads) +{ + uint64_t port_offloads = priv->dev->data->dev_conf.txmode.offloads; + uint64_t port_supp_offloads = mlx5_priv_get_tx_port_offloads(priv); + + /* There are no Tx offloads which are per queue. */ + if ((offloads & port_supp_offloads) != offloads) + return 0; + if ((port_offloads ^ offloads) & port_supp_offloads) + return 0; + return 1; +} + /** * DPDK callback to configure a TX queue. * @@ -143,6 +200,20 @@ mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, int ret = 0; priv_lock(priv); + /* + * Don't verify port offloads for application which + * use the old API. + */ + if (!!(conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) && + !priv_is_tx_queue_offloads_allowed(priv, conf->offloads)) { + ret = ENOTSUP; + ERROR("%p: Tx queue offloads 0x%" PRIx64 " don't match port " + "offloads 0x%" PRIx64 " or supported offloads 0x%" PRIx64, + (void *)dev, conf->offloads, + dev->data->dev_conf.txmode.offloads, + mlx5_priv_get_tx_port_offloads(priv)); + goto out; + } if (desc <= MLX5_TX_COMP_THRESH) { WARN("%p: number of descriptors requested for TX queue %u" " must be higher than MLX5_TX_COMP_THRESH, using" @@ -579,6 +650,7 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) unsigned int inline_max_packet_sz; eth_tx_burst_t tx_pkt_burst = priv_select_tx_function(priv, priv->dev); int is_empw_func = is_empw_burst_func(tx_pkt_burst); + int tso = !!(txq_ctrl->txq.offloads & DEV_TX_OFFLOAD_TCP_TSO); txq_inline = (config->txq_inline == MLX5_ARG_UNSET) ? 0 : config->txq_inline; @@ -603,8 +675,6 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) txq_ctrl->txq.max_inline = ((txq_inline + (RTE_CACHE_LINE_SIZE - 1)) / RTE_CACHE_LINE_SIZE); - /* TSO and MPS can't be enabled concurrently. */ - assert(!config->tso || !config->mps); if (is_empw_func) { /* To minimize the size of data set, avoid requesting * too large WQ. @@ -614,7 +684,7 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) inline_max_packet_sz) + (RTE_CACHE_LINE_SIZE - 1)) / RTE_CACHE_LINE_SIZE) * RTE_CACHE_LINE_SIZE; - } else if (config->tso) { + } else if (tso) { int inline_diff = txq_ctrl->txq.max_inline - max_tso_inline; @@ -652,7 +722,7 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) RTE_CACHE_LINE_SIZE; } } - if (config->tso) { + if (tso) { txq_ctrl->max_tso_header = max_tso_inline * RTE_CACHE_LINE_SIZE; txq_ctrl->txq.max_inline = RTE_MAX(txq_ctrl->txq.max_inline, max_tso_inline); @@ -692,7 +762,7 @@ mlx5_priv_txq_new(struct priv *priv, uint16_t idx, uint16_t desc, if (!tmpl) return NULL; assert(desc > MLX5_TX_COMP_THRESH); - tmpl->txq.flags = conf->txq_flags; + tmpl->txq.offloads = conf->offloads; tmpl->priv = priv; tmpl->socket = socket; tmpl->txq.elts_n = log2above(desc); -- 2.20.1