X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx4%2Fmlx4.c;h=d41552bd6889f8a96922c87c1d74a06fe76eeb08;hb=d7769c7c08cc08a9d1bc4e40b95524d9697707d9;hp=941176ca7ce020a367bfccb397065ae9bd80b5dc;hpb=2c960a5116d05cc7c93d00df707655d72d73d230;p=dpdk.git diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c index 941176ca7c..d41552bd68 100644 --- a/drivers/net/mlx4/mlx4.c +++ b/drivers/net/mlx4/mlx4.c @@ -1,8 +1,8 @@ /*- * BSD LICENSE * - * Copyright 2012-2015 6WIND S.A. - * Copyright 2012 Mellanox. + * Copyright 2012-2017 6WIND S.A. + * Copyright 2012-2017 Mellanox. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -58,22 +58,9 @@ #include #include -/* Verbs header. */ -/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */ -#ifdef PEDANTIC -#pragma GCC diagnostic ignored "-Wpedantic" -#endif -#include -#ifdef PEDANTIC -#pragma GCC diagnostic error "-Wpedantic" -#endif - -/* DPDK headers don't like -pedantic. */ -#ifdef PEDANTIC -#pragma GCC diagnostic ignored "-Wpedantic" -#endif #include #include +#include #include #include #include @@ -86,30 +73,16 @@ #include #include #include -#ifdef PEDANTIC -#pragma GCC diagnostic error "-Wpedantic" -#endif +#include +#include +#include /* Generated configuration header. */ #include "mlx4_autoconf.h" -/* PMD header. */ +/* PMD headers. */ #include "mlx4.h" - -/* Runtime logging through RTE_LOG() is enabled when not in debugging mode. - * Intermediate LOG_*() macros add the required end-of-line characters. */ -#ifndef NDEBUG -#define INFO(...) DEBUG(__VA_ARGS__) -#define WARN(...) DEBUG(__VA_ARGS__) -#define ERROR(...) DEBUG(__VA_ARGS__) -#else -#define LOG__(level, m, ...) \ - RTE_LOG(level, PMD, MLX4_DRIVER_NAME ": " m "%c", __VA_ARGS__) -#define LOG_(level, ...) LOG__(level, __VA_ARGS__, '\n') -#define INFO(...) LOG_(INFO, __VA_ARGS__) -#define WARN(...) LOG_(WARNING, __VA_ARGS__) -#define ERROR(...) LOG_(ERR, __VA_ARGS__) -#endif +#include "mlx4_flow.h" /* Convenience macros for accessing mbuf fields. */ #define NEXT(m) ((m)->next) @@ -137,164 +110,35 @@ typedef union { (((val) & (from)) / ((from) / (to))) : \ (((val) & (from)) * ((to) / (from)))) -struct mlx4_rxq_stats { - unsigned int idx; /**< Mapping index. */ -#ifdef MLX4_PMD_SOFT_COUNTERS - uint64_t ipackets; /**< Total of successfully received packets. */ - uint64_t ibytes; /**< Total of successfully received bytes. */ -#endif - uint64_t idropped; /**< Total of packets dropped when RX ring full. */ - uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */ -}; - -struct mlx4_txq_stats { - unsigned int idx; /**< Mapping index. */ -#ifdef MLX4_PMD_SOFT_COUNTERS - uint64_t opackets; /**< Total of successfully sent packets. */ - uint64_t obytes; /**< Total of successfully sent bytes. */ -#endif - uint64_t odropped; /**< Total of packets not sent when TX ring full. */ -}; - -/* RX element (scattered packets). */ -struct rxq_elt_sp { - struct ibv_recv_wr wr; /* Work Request. */ - struct ibv_sge sges[MLX4_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */ - struct rte_mbuf *bufs[MLX4_PMD_SGE_WR_N]; /* SGEs buffers. */ -}; +/* Local storage for secondary process data. */ +struct mlx4_secondary_data { + struct rte_eth_dev_data data; /* Local device data. */ + struct priv *primary_priv; /* Private structure from primary. */ + struct rte_eth_dev_data *shared_dev_data; /* Shared device data. */ + rte_spinlock_t lock; /* Port configuration lock. */ +} mlx4_secondary_data[RTE_MAX_ETHPORTS]; -/* RX element. */ -struct rxq_elt { - struct ibv_recv_wr wr; /* Work Request. */ - struct ibv_sge sge; /* Scatter/Gather Element. */ - /* mbuf pointer is derived from WR_ID(wr.wr_id).offset. */ +struct mlx4_conf { + uint8_t active_ports; }; -/* RX queue descriptor. */ -struct rxq { - struct priv *priv; /* Back pointer to private data. */ - struct rte_mempool *mp; /* Memory Pool for allocations. */ - struct ibv_mr *mr; /* Memory Region (for mp). */ - struct ibv_cq *cq; /* Completion Queue. */ - struct ibv_qp *qp; /* Queue Pair. */ - struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */ - struct ibv_exp_cq_family *if_cq; /* CQ interface. */ - /* - * Each VLAN ID requires a separate flow steering rule. - */ - BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES); - struct ibv_flow *mac_flow[MLX4_MAX_MAC_ADDRESSES][MLX4_MAX_VLAN_IDS]; - struct ibv_flow *promisc_flow; /* Promiscuous flow. */ - struct ibv_flow *allmulti_flow; /* Multicast flow. */ - unsigned int port_id; /* Port ID for incoming packets. */ - unsigned int elts_n; /* (*elts)[] length. */ - unsigned int elts_head; /* Current index in (*elts)[]. */ - union { - struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */ - struct rxq_elt (*no_sp)[]; /* RX elements. */ - } elts; - unsigned int sp:1; /* Use scattered RX elements. */ - unsigned int csum:1; /* Enable checksum offloading. */ - unsigned int csum_l2tun:1; /* Same for L2 tunnels. */ - struct mlx4_rxq_stats stats; /* RX queue counters. */ - unsigned int socket; /* CPU socket ID for allocations. */ - struct ibv_exp_res_domain *rd; /* Resource Domain. */ +/* Available parameters list. */ +const char *pmd_mlx4_init_params[] = { + MLX4_PMD_PORT_KVARG, + NULL, }; -/* TX element. */ -struct txq_elt { - struct rte_mbuf *buf; -}; +static int +mlx4_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx); -/* Linear buffer type. It is used when transmitting buffers with too many - * segments that do not fit the hardware queue (see max_send_sge). - * Extra segments are copied (linearized) in such buffers, replacing the - * last SGE during TX. - * The size is arbitrary but large enough to hold a jumbo frame with - * 8 segments considering mbuf.buf_len is about 2048 bytes. */ -typedef uint8_t linear_t[16384]; - -/* TX queue descriptor. */ -struct txq { - struct priv *priv; /* Back pointer to private data. */ - struct { - const struct rte_mempool *mp; /* Cached Memory Pool. */ - struct ibv_mr *mr; /* Memory Region (for mp). */ - uint32_t lkey; /* mr->lkey */ - } mp2mr[MLX4_PMD_TX_MP_CACHE]; /* MP to MR translation table. */ - struct ibv_cq *cq; /* Completion Queue. */ - struct ibv_qp *qp; /* Queue Pair. */ - struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */ - struct ibv_exp_cq_family *if_cq; /* CQ interface. */ -#if MLX4_PMD_MAX_INLINE > 0 - uint32_t max_inline; /* Max inline send size <= MLX4_PMD_MAX_INLINE. */ -#endif - unsigned int elts_n; /* (*elts)[] length. */ - struct txq_elt (*elts)[]; /* TX elements. */ - unsigned int elts_head; /* Current index in (*elts)[]. */ - unsigned int elts_tail; /* First element awaiting completion. */ - unsigned int elts_comp; /* Number of completion requests. */ - unsigned int elts_comp_cd; /* Countdown for next completion request. */ - unsigned int elts_comp_cd_init; /* Initial value for countdown. */ - struct mlx4_txq_stats stats; /* TX queue counters. */ - linear_t (*elts_linear)[]; /* Linearized buffers. */ - struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */ - unsigned int socket; /* CPU socket ID for allocations. */ - struct ibv_exp_res_domain *rd; /* Resource Domain. */ -}; +static int +mlx4_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx); -struct priv { - struct rte_eth_dev *dev; /* Ethernet device. */ - struct ibv_context *ctx; /* Verbs context. */ - struct ibv_device_attr device_attr; /* Device properties. */ - struct ibv_pd *pd; /* Protection Domain. */ - /* - * MAC addresses array and configuration bit-field. - * An extra entry that cannot be modified by the DPDK is reserved - * for broadcast frames (destination MAC address ff:ff:ff:ff:ff:ff). - */ - struct ether_addr mac[MLX4_MAX_MAC_ADDRESSES]; - BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES); - /* VLAN filters. */ - struct { - unsigned int enabled:1; /* If enabled. */ - unsigned int id:12; /* VLAN ID (0-4095). */ - } vlan_filter[MLX4_MAX_VLAN_IDS]; /* VLAN filters table. */ - /* Device properties. */ - uint16_t mtu; /* Configured MTU. */ - uint8_t port; /* Physical port number. */ - unsigned int started:1; /* Device started, flows enabled. */ - unsigned int promisc:1; /* Device in promiscuous mode. */ - unsigned int allmulti:1; /* Device receives all multicast packets. */ - unsigned int hw_qpg:1; /* QP groups are supported. */ - unsigned int hw_tss:1; /* TSS is supported. */ - unsigned int hw_rss:1; /* RSS is supported. */ - unsigned int hw_csum:1; /* Checksum offload is supported. */ - unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */ - unsigned int rss:1; /* RSS is enabled. */ - unsigned int vf:1; /* This is a VF device. */ - unsigned int pending_alarm:1; /* An alarm is pending. */ -#ifdef INLINE_RECV - unsigned int inl_recv_size; /* Inline recv size */ -#endif - unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */ - /* RX/TX queues. */ - struct rxq rxq_parent; /* Parent queue when RSS is enabled. */ - unsigned int rxqs_n; /* RX queues array size. */ - unsigned int txqs_n; /* TX queues array size. */ - struct rxq *(*rxqs)[]; /* RX queues. */ - struct txq *(*txqs)[]; /* TX queues. */ - struct rte_intr_handle intr_handle; /* Interrupt handler. */ - rte_spinlock_t lock; /* Lock for control functions. */ -}; +static int +priv_rx_intr_vec_enable(struct priv *priv); -/* Local storage for secondary process data. */ -struct mlx4_secondary_data { - struct rte_eth_dev_data data; /* Local device data. */ - struct priv *primary_priv; /* Private structure from primary. */ - struct rte_eth_dev_data *shared_dev_data; /* Shared device data. */ - rte_spinlock_t lock; /* Port configuration lock. */ -} mlx4_secondary_data[RTE_MAX_ETHPORTS]; +static void +priv_rx_intr_vec_disable(struct priv *priv); /** * Check if running as a secondary process. @@ -335,8 +179,7 @@ mlx4_get_priv(struct rte_eth_dev *dev) * @param priv * Pointer to private structure. */ -static void -priv_lock(struct priv *priv) +void priv_lock(struct priv *priv) { rte_spinlock_lock(&priv->lock); } @@ -347,8 +190,7 @@ priv_lock(struct priv *priv) * @param priv * Pointer to private structure. */ -static void -priv_unlock(struct priv *priv) +void priv_unlock(struct priv *priv) { rte_spinlock_unlock(&priv->lock); } @@ -704,12 +546,92 @@ txq_cleanup(struct txq *txq); static int rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, - unsigned int socket, int inactive, const struct rte_eth_rxconf *conf, - struct rte_mempool *mp); + unsigned int socket, int inactive, + const struct rte_eth_rxconf *conf, + struct rte_mempool *mp, int children_n, + struct rxq *rxq_parent); static void rxq_cleanup(struct rxq *rxq); +/** + * Create RSS parent queue. + * + * The new parent is inserted in front of the list in the private structure. + * + * @param priv + * Pointer to private structure. + * @param queues + * Queues indices array, if NULL use all Rx queues. + * @param children_n + * The number of entries in queues[]. + * + * @return + * Pointer to a parent rxq structure, NULL on failure. + */ +struct rxq * +priv_parent_create(struct priv *priv, + uint16_t queues[], + uint16_t children_n) +{ + int ret; + uint16_t i; + struct rxq *parent; + + parent = rte_zmalloc("parent queue", + sizeof(*parent), + RTE_CACHE_LINE_SIZE); + if (!parent) { + ERROR("cannot allocate memory for RSS parent queue"); + return NULL; + } + ret = rxq_setup(priv->dev, parent, 0, 0, 0, + NULL, NULL, children_n, NULL); + if (ret) { + rte_free(parent); + return NULL; + } + parent->rss.queues_n = children_n; + if (queues) { + for (i = 0; i < children_n; ++i) + parent->rss.queues[i] = queues[i]; + } else { + /* the default RSS ring case */ + assert(priv->rxqs_n == children_n); + for (i = 0; i < priv->rxqs_n; ++i) + parent->rss.queues[i] = i; + } + LIST_INSERT_HEAD(&priv->parents, parent, next); + return parent; +} + +/** + * Clean up RX queue parent structure. + * + * @param parent + * RX queue parent structure. + */ +void +rxq_parent_cleanup(struct rxq *parent) +{ + LIST_REMOVE(parent, next); + rxq_cleanup(parent); + rte_free(parent); +} + +/** + * Clean up parent structures from the parent list. + * + * @param priv + * Pointer to private structure. + */ +static void +priv_parent_list_cleanup(struct priv *priv) +{ + while (!LIST_EMPTY(&priv->parents)) + rxq_parent_cleanup(LIST_FIRST(&priv->parents)); +} + /** * Ethernet device configuration. * @@ -729,7 +651,6 @@ dev_configure(struct rte_eth_dev *dev) unsigned int rxqs_n = dev->data->nb_rx_queues; unsigned int txqs_n = dev->data->nb_tx_queues; unsigned int tmp; - int ret; priv->rxqs = (void *)dev->data->rx_queues; priv->txqs = (void *)dev->data->tx_queues; @@ -740,7 +661,7 @@ dev_configure(struct rte_eth_dev *dev) } if (rxqs_n == priv->rxqs_n) return 0; - if (!rte_is_power_of_2(rxqs_n)) { + if (!rte_is_power_of_2(rxqs_n) && !priv->isolated) { unsigned n_active; n_active = rte_align32pow2(rxqs_n + 1) >> 1; @@ -759,7 +680,7 @@ dev_configure(struct rte_eth_dev *dev) for (i = 0; (i != priv->rxqs_n); ++i) if ((*priv->rxqs)[i] != NULL) return EINVAL; - rxq_cleanup(&priv->rxq_parent); + priv_parent_list_cleanup(priv); priv->rss = 0; priv->rxqs_n = 0; } @@ -784,14 +705,14 @@ dev_configure(struct rte_eth_dev *dev) priv->rss = 1; tmp = priv->rxqs_n; priv->rxqs_n = rxqs_n; - ret = rxq_setup(dev, &priv->rxq_parent, 0, 0, 0, NULL, NULL); - if (!ret) + if (priv->isolated) + return 0; + if (priv_parent_create(priv, NULL, priv->rxqs_n)) return 0; /* Failure, rollback. */ priv->rss = 0; priv->rxqs_n = tmp; - assert(ret > 0); - return ret; + return ENOMEM; } /** @@ -1269,7 +1190,7 @@ static int mlx4_check_mempool(struct rte_mempool *mp, uintptr_t *start, /* For best performance, this function should not be inlined. */ static struct ibv_mr *mlx4_mp2mr(struct ibv_pd *, struct rte_mempool *) - __attribute__((noinline)); + __rte_noinline; /** * Register mempool as a memory region. @@ -2526,6 +2447,7 @@ rxq_add_flow(struct rxq *rxq, unsigned int mac_index, unsigned int vlan_index) assert(((uint8_t *)attr + sizeof(*attr)) == (uint8_t *)spec); *attr = (struct ibv_flow_attr){ .type = IBV_FLOW_ATTR_NORMAL, + .priority = 3, .num_of_specs = 1, .port = priv->port, .flags = 0 @@ -2669,11 +2591,12 @@ priv_mac_addr_del(struct priv *priv, unsigned int mac_index) { unsigned int i; + assert(!priv->isolated); assert(mac_index < elemof(priv->mac)); if (!BITFIELD_ISSET(priv->mac_configured, mac_index)) return; if (priv->rss) { - rxq_mac_addr_del(&priv->rxq_parent, mac_index); + rxq_mac_addr_del(LIST_FIRST(&priv->parents), mac_index); goto end; } for (i = 0; (i != priv->dev->data->nb_rx_queues); ++i) @@ -2740,7 +2663,7 @@ priv_mac_addr_add(struct priv *priv, unsigned int mac_index, goto end; } if (priv->rss) { - ret = rxq_mac_addr_add(&priv->rxq_parent, mac_index); + ret = rxq_mac_addr_add(LIST_FIRST(&priv->parents), mac_index); if (ret) return ret; goto end; @@ -2918,14 +2841,17 @@ rxq_cleanup(struct rxq *rxq) rxq->if_cq, ¶ms)); } - if (rxq->qp != NULL) { + if (rxq->qp != NULL && !rxq->priv->isolated) { rxq_promiscuous_disable(rxq); rxq_allmulticast_disable(rxq); rxq_mac_addrs_del(rxq); - claim_zero(ibv_destroy_qp(rxq->qp)); } + if (rxq->qp != NULL) + claim_zero(ibv_destroy_qp(rxq->qp)); if (rxq->cq != NULL) claim_zero(ibv_destroy_cq(rxq->cq)); + if (rxq->channel != NULL) + claim_zero(ibv_destroy_comp_channel(rxq->channel)); if (rxq->rd != NULL) { struct ibv_exp_destroy_res_domain_attr attr = { .comp_mask = 0, @@ -2961,19 +2887,25 @@ rxq_cq_to_pkt_type(uint32_t flags) if (flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) pkt_type = TRANSPOSE(flags, - IBV_EXP_CQ_RX_OUTER_IPV4_PACKET, RTE_PTYPE_L3_IPV4) | + IBV_EXP_CQ_RX_OUTER_IPV4_PACKET, + RTE_PTYPE_L3_IPV4_EXT_UNKNOWN) | TRANSPOSE(flags, - IBV_EXP_CQ_RX_OUTER_IPV6_PACKET, RTE_PTYPE_L3_IPV6) | + IBV_EXP_CQ_RX_OUTER_IPV6_PACKET, + RTE_PTYPE_L3_IPV6_EXT_UNKNOWN) | TRANSPOSE(flags, - IBV_EXP_CQ_RX_IPV4_PACKET, RTE_PTYPE_INNER_L3_IPV4) | + IBV_EXP_CQ_RX_IPV4_PACKET, + RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN) | TRANSPOSE(flags, - IBV_EXP_CQ_RX_IPV6_PACKET, RTE_PTYPE_INNER_L3_IPV6); + IBV_EXP_CQ_RX_IPV6_PACKET, + RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN); else pkt_type = TRANSPOSE(flags, - IBV_EXP_CQ_RX_IPV4_PACKET, RTE_PTYPE_L3_IPV4) | + IBV_EXP_CQ_RX_IPV4_PACKET, + RTE_PTYPE_L3_IPV4_EXT_UNKNOWN) | TRANSPOSE(flags, - IBV_EXP_CQ_RX_IPV6_PACKET, RTE_PTYPE_L3_IPV6); + IBV_EXP_CQ_RX_IPV6_PACKET, + RTE_PTYPE_L3_IPV6_EXT_UNKNOWN); return pkt_type; } @@ -3151,6 +3083,13 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) NB_SEGS(rep) = 0x2a; PORT(rep) = 0x2a; rep->ol_flags = -1; + /* + * Clear special flags in mbuf to avoid + * crashing while freeing. + */ + rep->ol_flags &= + ~(uint64_t)(IND_ATTACHED_MBUF | + CTRL_MBUF_FLAG); #endif assert(rep->buf_len == seg->buf_len); /* Reconfigure sge to use rep instead of seg. */ @@ -3334,6 +3273,8 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) /* Increase out of memory counters. */ ++rxq->stats.rx_nombuf; ++rxq->priv->dev->data->rx_mbuf_alloc_failed; + /* Add SGE to array for repost. */ + sges[i] = elt->sge; goto repost; } @@ -3492,15 +3433,18 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc, * Completion queue to associate with QP. * @param desc * Number of descriptors in QP (hint only). - * @param parent - * If nonzero, create a parent QP, otherwise a child. + * @param children_n + * If nonzero, a number of children for parent QP and zero for a child. + * @param rxq_parent + * Pointer for a parent in a child case, NULL otherwise. * * @return * QP pointer or NULL in case of error. */ static struct ibv_qp * rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc, - int parent, struct ibv_exp_res_domain *rd) + int children_n, struct ibv_exp_res_domain *rd, + struct rxq *rxq_parent) { struct ibv_exp_qp_init_attr attr = { /* CQ to be associated with the send queue. */ @@ -3530,16 +3474,16 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc, attr.max_inl_recv = priv->inl_recv_size, attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV; #endif - if (parent) { + if (children_n > 0) { attr.qpg.qpg_type = IBV_EXP_QPG_PARENT; /* TSS isn't necessary. */ attr.qpg.parent_attrib.tss_child_count = 0; attr.qpg.parent_attrib.rss_child_count = - rte_align32pow2(priv->rxqs_n + 1) >> 1; + rte_align32pow2(children_n + 1) >> 1; DEBUG("initializing parent RSS queue"); } else { attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX; - attr.qpg.qpg_parent = priv->rxq_parent.qp; + attr.qpg.qpg_parent = rxq_parent->qp; DEBUG("initializing child RSS queue"); } return ibv_exp_create_qp(priv->ctx, &attr); @@ -3575,13 +3519,7 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) struct ibv_recv_wr *bad_wr; unsigned int mb_len; int err; - int parent = (rxq == &priv->rxq_parent); - if (parent) { - ERROR("%p: cannot rehash parent queue %p", - (void *)dev, (void *)rxq); - return EINVAL; - } mb_len = rte_pktmbuf_data_room_size(rxq->mp); DEBUG("%p: rehashing queue %p", (void *)dev, (void *)rxq); /* Number of descriptors and mbufs currently allocated. */ @@ -3598,7 +3536,7 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) } /* Enable scattered packets support for this queue if necessary. */ assert(mb_len >= RTE_PKTMBUF_HEADROOM); - if ((dev->data->dev_conf.rxmode.jumbo_frame) && + if (dev->data->dev_conf.rxmode.enable_scatter && (dev->data->dev_conf.rxmode.max_rx_pkt_len > (mb_len - RTE_PKTMBUF_HEADROOM))) { tmpl.sp = 1; @@ -3613,7 +3551,7 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) return 0; } /* Remove attached flows if RSS is disabled (no parent queue). */ - if (!priv->rss) { + if (!priv->rss && !priv->isolated) { rxq_allmulticast_disable(&tmpl); rxq_promiscuous_disable(&tmpl); rxq_mac_addrs_del(&tmpl); @@ -3626,6 +3564,8 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) } /* From now on, any failure will render the queue unusable. * Reinitialize QP. */ + if (!tmpl.qp) + goto skip_init; mod = (struct ibv_exp_qp_attr){ .qp_state = IBV_QPS_RESET }; err = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE); if (err) { @@ -3633,12 +3573,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) assert(err > 0); return err; } - err = ibv_resize_cq(tmpl.cq, desc_n); - if (err) { - ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err)); - assert(err > 0); - return err; - } mod = (struct ibv_exp_qp_attr){ /* Move the QP to this state. */ .qp_state = IBV_QPS_INIT, @@ -3647,9 +3581,6 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) }; err = ibv_exp_modify_qp(tmpl.qp, &mod, (IBV_EXP_QP_STATE | -#ifdef RSS_SUPPORT - (parent ? IBV_EXP_QP_GROUP_RSS : 0) | -#endif /* RSS_SUPPORT */ IBV_EXP_QP_PORT)); if (err) { ERROR("%p: QP state to IBV_QPS_INIT failed: %s", @@ -3657,8 +3588,15 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) assert(err > 0); return err; }; +skip_init: + err = ibv_resize_cq(tmpl.cq, desc_n); + if (err) { + ERROR("%p: cannot resize CQ: %s", (void *)dev, strerror(err)); + assert(err > 0); + return err; + } /* Reconfigure flows. Do not care for errors. */ - if (!priv->rss) { + if (!priv->rss && !priv->isolated) { rxq_mac_addrs_add(&tmpl); if (priv->promisc) rxq_promiscuous_enable(&tmpl); @@ -3724,6 +3662,8 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) rxq->elts_n = 0; rte_free(rxq->elts.sp); rxq->elts.sp = NULL; + if (!tmpl.qp) + goto skip_rtr; /* Post WRs. */ err = ibv_post_recv(tmpl.qp, (tmpl.sp ? @@ -3750,6 +3690,116 @@ skip_rtr: return err; } +/** + * Create verbs QP resources associated with a rxq. + * + * @param rxq + * Pointer to RX queue structure. + * @param desc + * Number of descriptors to configure in queue. + * @param inactive + * If true, the queue is disabled because its index is higher or + * equal to the real number of queues, which must be a power of 2. + * @param children_n + * The number of children in a parent case, zero for a child. + * @param rxq_parent + * The pointer to a parent RX structure for a child in RSS case, + * NULL for parent. + * + * @return + * 0 on success, errno value on failure. + */ +int +rxq_create_qp(struct rxq *rxq, + uint16_t desc, + int inactive, + int children_n, + struct rxq *rxq_parent) +{ + int ret; + struct ibv_exp_qp_attr mod; + struct ibv_exp_query_intf_params params; + enum ibv_exp_query_intf_status status; + struct ibv_recv_wr *bad_wr; + int parent = (children_n > 0); + struct priv *priv = rxq->priv; + +#ifdef RSS_SUPPORT + if (priv->rss && !inactive && (rxq_parent || parent)) + rxq->qp = rxq_setup_qp_rss(priv, rxq->cq, desc, + children_n, rxq->rd, + rxq_parent); + else +#endif /* RSS_SUPPORT */ + rxq->qp = rxq_setup_qp(priv, rxq->cq, desc, rxq->rd); + if (rxq->qp == NULL) { + ret = (errno ? errno : EINVAL); + ERROR("QP creation failure: %s", + strerror(ret)); + return ret; + } + mod = (struct ibv_exp_qp_attr){ + /* Move the QP to this state. */ + .qp_state = IBV_QPS_INIT, + /* Primary port number. */ + .port_num = priv->port + }; + ret = ibv_exp_modify_qp(rxq->qp, &mod, + (IBV_EXP_QP_STATE | +#ifdef RSS_SUPPORT + (parent ? IBV_EXP_QP_GROUP_RSS : 0) | +#endif /* RSS_SUPPORT */ + IBV_EXP_QP_PORT)); + if (ret) { + ERROR("QP state to IBV_QPS_INIT failed: %s", + strerror(ret)); + return ret; + } + if (!priv->isolated && (parent || !priv->rss)) { + /* Configure MAC and broadcast addresses. */ + ret = rxq_mac_addrs_add(rxq); + if (ret) { + ERROR("QP flow attachment failed: %s", + strerror(ret)); + return ret; + } + } + if (!parent) { + ret = ibv_post_recv(rxq->qp, + (rxq->sp ? + &(*rxq->elts.sp)[0].wr : + &(*rxq->elts.no_sp)[0].wr), + &bad_wr); + if (ret) { + ERROR("ibv_post_recv() failed for WR %p: %s", + (void *)bad_wr, + strerror(ret)); + return ret; + } + } + mod = (struct ibv_exp_qp_attr){ + .qp_state = IBV_QPS_RTR + }; + ret = ibv_exp_modify_qp(rxq->qp, &mod, IBV_EXP_QP_STATE); + if (ret) { + ERROR("QP state to IBV_QPS_RTR failed: %s", + strerror(ret)); + return ret; + } + params = (struct ibv_exp_query_intf_params){ + .intf_scope = IBV_EXP_INTF_GLOBAL, + .intf = IBV_EXP_INTF_QP_BURST, + .obj = rxq->qp, + }; + rxq->if_qp = ibv_exp_query_intf(priv->ctx, ¶ms, &status); + if (rxq->if_qp == NULL) { + ERROR("QP interface family query failed with status %d", + status); + return errno; + } + return 0; +} + /** * Configure a RX queue. * @@ -3768,14 +3818,21 @@ skip_rtr: * Thresholds parameters. * @param mp * Memory pool for buffer allocations. + * @param children_n + * The number of children in a parent case, zero for a child. + * @param rxq_parent + * The pointer to a parent RX structure (or NULL) in a child case, + * NULL for parent. * * @return * 0 on success, errno value on failure. */ static int rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, - unsigned int socket, int inactive, const struct rte_eth_rxconf *conf, - struct rte_mempool *mp) + unsigned int socket, int inactive, + const struct rte_eth_rxconf *conf, + struct rte_mempool *mp, int children_n, + struct rxq *rxq_parent) { struct priv *priv = dev->data->dev_private; struct rxq tmpl = { @@ -3783,17 +3840,15 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, .mp = mp, .socket = socket }; - struct ibv_exp_qp_attr mod; union { struct ibv_exp_query_intf_params params; struct ibv_exp_cq_init_attr cq; struct ibv_exp_res_domain_init_attr rd; } attr; enum ibv_exp_query_intf_status status; - struct ibv_recv_wr *bad_wr; unsigned int mb_len; int ret = 0; - int parent = (rxq == &priv->rxq_parent); + int parent = (children_n > 0); (void)conf; /* Thresholds configuration (ignored). */ /* @@ -3820,11 +3875,19 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum; /* Enable scattered packets support for this queue if necessary. */ assert(mb_len >= RTE_PKTMBUF_HEADROOM); - if ((dev->data->dev_conf.rxmode.jumbo_frame) && - (dev->data->dev_conf.rxmode.max_rx_pkt_len > - (mb_len - RTE_PKTMBUF_HEADROOM))) { + if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= + (mb_len - RTE_PKTMBUF_HEADROOM)) { + tmpl.sp = 0; + } else if (dev->data->dev_conf.rxmode.enable_scatter) { tmpl.sp = 1; desc /= MLX4_PMD_SGE_WR_N; + } else { + WARN("%p: the requested maximum Rx packet size (%u) is" + " larger than a single mbuf (%u) and scattered" + " mode has not been requested", + (void *)dev, + dev->data->dev_conf.rxmode.max_rx_pkt_len, + mb_len - RTE_PKTMBUF_HEADROOM); } DEBUG("%p: %s scattered packets support (%u WRs)", (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc); @@ -3850,11 +3913,22 @@ skip_mr: (void *)dev, strerror(ret)); goto error; } + if (dev->data->dev_conf.intr_conf.rxq) { + tmpl.channel = ibv_create_comp_channel(priv->ctx); + if (tmpl.channel == NULL) { + ret = ENOMEM; + ERROR("%p: Rx interrupt completion channel creation" + " failure: %s", + (void *)dev, strerror(ret)); + goto error; + } + } attr.cq = (struct ibv_exp_cq_init_attr){ .comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN, .res_domain = tmpl.rd, }; - tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq); + tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, tmpl.channel, 0, + &attr.cq); if (tmpl.cq == NULL) { ret = ENOMEM; ERROR("%p: CQ creation failure: %s", @@ -3865,45 +3939,6 @@ skip_mr: priv->device_attr.max_qp_wr); DEBUG("priv->device_attr.max_sge is %d", priv->device_attr.max_sge); -#ifdef RSS_SUPPORT - if (priv->rss && !inactive) - tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent, - tmpl.rd); - else -#endif /* RSS_SUPPORT */ - tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd); - if (tmpl.qp == NULL) { - ret = (errno ? errno : EINVAL); - ERROR("%p: QP creation failure: %s", - (void *)dev, strerror(ret)); - goto error; - } - mod = (struct ibv_exp_qp_attr){ - /* Move the QP to this state. */ - .qp_state = IBV_QPS_INIT, - /* Primary port number. */ - .port_num = priv->port - }; - ret = ibv_exp_modify_qp(tmpl.qp, &mod, - (IBV_EXP_QP_STATE | -#ifdef RSS_SUPPORT - (parent ? IBV_EXP_QP_GROUP_RSS : 0) | -#endif /* RSS_SUPPORT */ - IBV_EXP_QP_PORT)); - if (ret) { - ERROR("%p: QP state to IBV_QPS_INIT failed: %s", - (void *)dev, strerror(ret)); - goto error; - } - if ((parent) || (!priv->rss)) { - /* Configure MAC and broadcast addresses. */ - ret = rxq_mac_addrs_add(&tmpl); - if (ret) { - ERROR("%p: QP flow attachment failed: %s", - (void *)dev, strerror(ret)); - goto error; - } - } /* Allocate descriptors for RX queues, except for the RSS parent. */ if (parent) goto skip_alloc; @@ -3914,29 +3949,14 @@ skip_mr: if (ret) { ERROR("%p: RXQ allocation failed: %s", (void *)dev, strerror(ret)); - goto error; - } - ret = ibv_post_recv(tmpl.qp, - (tmpl.sp ? - &(*tmpl.elts.sp)[0].wr : - &(*tmpl.elts.no_sp)[0].wr), - &bad_wr); - if (ret) { - ERROR("%p: ibv_post_recv() failed for WR %p: %s", - (void *)dev, - (void *)bad_wr, - strerror(ret)); - goto error; + return ret; } skip_alloc: - mod = (struct ibv_exp_qp_attr){ - .qp_state = IBV_QPS_RTR - }; - ret = ibv_exp_modify_qp(tmpl.qp, &mod, IBV_EXP_QP_STATE); - if (ret) { - ERROR("%p: QP state to IBV_QPS_RTR failed: %s", - (void *)dev, strerror(ret)); - goto error; + if (parent || rxq_parent || !priv->rss) { + ret = rxq_create_qp(&tmpl, desc, inactive, + children_n, rxq_parent); + if (ret) + goto error; } /* Save port ID. */ tmpl.port_id = dev->data->port_id; @@ -3948,21 +3968,11 @@ skip_alloc: }; tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status); if (tmpl.if_cq == NULL) { + ret = EINVAL; ERROR("%p: CQ interface family query failed with status %d", (void *)dev, status); goto error; } - attr.params = (struct ibv_exp_query_intf_params){ - .intf_scope = IBV_EXP_INTF_GLOBAL, - .intf = IBV_EXP_INTF_QP_BURST, - .obj = tmpl.qp, - }; - tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status); - if (tmpl.if_qp == NULL) { - ERROR("%p: QP interface family query failed with status %d", - (void *)dev, status); - goto error; - } /* Clean up rxq in case we're reinitializing it. */ DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq); rxq_cleanup(rxq); @@ -4000,6 +4010,7 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, unsigned int socket, const struct rte_eth_rxconf *conf, struct rte_mempool *mp) { + struct rxq *parent; struct priv *priv = dev->data->dev_private; struct rxq *rxq = (*priv->rxqs)[idx]; int inactive = 0; @@ -4034,9 +4045,16 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, return -ENOMEM; } } - if (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1) - inactive = 1; - ret = rxq_setup(dev, rxq, desc, socket, inactive, conf, mp); + if (priv->rss && !priv->isolated) { + /* The list consists of the single default one. */ + parent = LIST_FIRST(&priv->parents); + if (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1) + inactive = 1; + } else { + parent = NULL; + } + ret = rxq_setup(dev, rxq, desc, socket, + inactive, conf, mp, 0, parent); if (ret) rte_free(rxq); else { @@ -4073,7 +4091,6 @@ mlx4_rx_queue_release(void *dpdk_rxq) return; priv = rxq->priv; priv_lock(priv); - assert(rxq != &priv->rxq_parent); for (i = 0; (i != priv->rxqs_n); ++i) if ((*priv->rxqs)[i] == rxq) { DEBUG("%p: removing RX queue %p from list", @@ -4086,9 +4103,15 @@ mlx4_rx_queue_release(void *dpdk_rxq) priv_unlock(priv); } -static void +static int priv_dev_interrupt_handler_install(struct priv *, struct rte_eth_dev *); +static int +priv_dev_removal_interrupt_handler_install(struct priv *, struct rte_eth_dev *); + +static int +priv_dev_link_interrupt_handler_install(struct priv *, struct rte_eth_dev *); + /** * DPDK callback to start the device. * @@ -4107,6 +4130,7 @@ mlx4_dev_start(struct rte_eth_dev *dev) unsigned int i = 0; unsigned int r; struct rxq *rxq; + int ret; if (mlx4_is_secondary()) return -E_RTE_SECONDARY; @@ -4117,8 +4141,11 @@ mlx4_dev_start(struct rte_eth_dev *dev) } DEBUG("%p: attaching configured flows to all RX queues", (void *)dev); priv->started = 1; - if (priv->rss) { - rxq = &priv->rxq_parent; + if (priv->isolated) { + rxq = NULL; + r = 1; + } else if (priv->rss) { + rxq = LIST_FIRST(&priv->parents); r = 1; } else { rxq = (*priv->rxqs)[0]; @@ -4126,8 +4153,6 @@ mlx4_dev_start(struct rte_eth_dev *dev) } /* Iterate only once when RSS is enabled. */ do { - int ret; - /* Ignore nonexistent RX queues. */ if (rxq == NULL) continue; @@ -4140,22 +4165,47 @@ mlx4_dev_start(struct rte_eth_dev *dev) continue; WARN("%p: QP flow attachment failed: %s", (void *)dev, strerror(ret)); - /* Rollback. */ - while (i != 0) { - rxq = (*priv->rxqs)[--i]; - if (rxq != NULL) { - rxq_allmulticast_disable(rxq); - rxq_promiscuous_disable(rxq); - rxq_mac_addrs_del(rxq); - } - } - priv->started = 0; - priv_unlock(priv); - return -ret; + goto err; } while ((--r) && ((rxq = (*priv->rxqs)[++i]), i)); - priv_dev_interrupt_handler_install(priv, dev); + ret = priv_dev_link_interrupt_handler_install(priv, dev); + if (ret) { + ERROR("%p: LSC handler install failed", + (void *)dev); + goto err; + } + ret = priv_dev_removal_interrupt_handler_install(priv, dev); + if (ret) { + ERROR("%p: RMV handler install failed", + (void *)dev); + goto err; + } + ret = priv_rx_intr_vec_enable(priv); + if (ret) { + ERROR("%p: Rx interrupt vector creation failed", + (void *)dev); + goto err; + } + ret = mlx4_priv_flow_start(priv); + if (ret) { + ERROR("%p: flow start failed: %s", + (void *)dev, strerror(ret)); + goto err; + } priv_unlock(priv); return 0; +err: + /* Rollback. */ + while (i != 0) { + rxq = (*priv->rxqs)[i--]; + if (rxq != NULL) { + rxq_allmulticast_disable(rxq); + rxq_promiscuous_disable(rxq); + rxq_mac_addrs_del(rxq); + } + } + priv->started = 0; + priv_unlock(priv); + return -ret; } /** @@ -4183,13 +4233,17 @@ mlx4_dev_stop(struct rte_eth_dev *dev) } DEBUG("%p: detaching flows from all RX queues", (void *)dev); priv->started = 0; - if (priv->rss) { - rxq = &priv->rxq_parent; + if (priv->isolated) { + rxq = NULL; + r = 1; + } else if (priv->rss) { + rxq = LIST_FIRST(&priv->parents); r = 1; } else { rxq = (*priv->rxqs)[0]; r = priv->rxqs_n; } + mlx4_priv_flow_stop(priv); /* Iterate only once when RSS is enabled. */ do { /* Ignore nonexistent RX queues. */ @@ -4252,9 +4306,16 @@ removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) return 0; } -static void +static int priv_dev_interrupt_handler_uninstall(struct priv *, struct rte_eth_dev *); +static int +priv_dev_removal_interrupt_handler_uninstall(struct priv *, + struct rte_eth_dev *); + +static int +priv_dev_link_interrupt_handler_uninstall(struct priv *, struct rte_eth_dev *); + /** * DPDK callback to close the device. * @@ -4310,14 +4371,16 @@ mlx4_dev_close(struct rte_eth_dev *dev) priv->txqs = NULL; } if (priv->rss) - rxq_cleanup(&priv->rxq_parent); + priv_parent_list_cleanup(priv); if (priv->pd != NULL) { assert(priv->ctx != NULL); claim_zero(ibv_dealloc_pd(priv->pd)); claim_zero(ibv_close_device(priv->ctx)); } else assert(priv->ctx == NULL); - priv_dev_interrupt_handler_uninstall(priv, dev); + priv_dev_removal_interrupt_handler_uninstall(priv, dev); + priv_dev_link_interrupt_handler_uninstall(priv, dev); + priv_rx_intr_vec_disable(priv); priv_unlock(priv); memset(priv, 0, sizeof(*priv)); } @@ -4421,7 +4484,7 @@ mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) unsigned int max; char ifname[IF_NAMESIZE]; - info->pci_dev = RTE_DEV_TO_PCI(dev->device); + info->pci_dev = RTE_ETH_DEV_TO_PCI(dev); if (priv == NULL) return; @@ -4602,6 +4665,8 @@ mlx4_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index) if (mlx4_is_secondary()) return; priv_lock(priv); + if (priv->isolated) + goto end; DEBUG("%p: removing MAC address from index %" PRIu32, (void *)dev, index); /* Last array entry is reserved for broadcast. */ @@ -4624,26 +4689,36 @@ end: * @param vmdq * VMDq pool index to associate address with (ignored). */ -static void +static int mlx4_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr, uint32_t index, uint32_t vmdq) { struct priv *priv = dev->data->dev_private; + int re; if (mlx4_is_secondary()) - return; + return -ENOTSUP; (void)vmdq; priv_lock(priv); + if (priv->isolated) { + DEBUG("%p: cannot add MAC address, " + "device is in isolated mode", (void *)dev); + re = EPERM; + goto end; + } DEBUG("%p: adding MAC address at index %" PRIu32, (void *)dev, index); /* Last array entry is reserved for broadcast. */ - if (index >= (elemof(priv->mac) - 1)) + if (index >= (elemof(priv->mac) - 1)) { + re = EINVAL; goto end; - priv_mac_addr_add(priv, index, - (const uint8_t (*)[ETHER_ADDR_LEN]) - mac_addr->addr_bytes); + } + re = priv_mac_addr_add(priv, index, + (const uint8_t (*)[ETHER_ADDR_LEN]) + mac_addr->addr_bytes); end: priv_unlock(priv); + return -re; } /** @@ -4678,6 +4753,12 @@ mlx4_promiscuous_enable(struct rte_eth_dev *dev) if (mlx4_is_secondary()) return; priv_lock(priv); + if (priv->isolated) { + DEBUG("%p: cannot enable promiscuous, " + "device is in isolated mode", (void *)dev); + priv_unlock(priv); + return; + } if (priv->promisc) { priv_unlock(priv); return; @@ -4686,7 +4767,7 @@ mlx4_promiscuous_enable(struct rte_eth_dev *dev) if (!priv->started) goto end; if (priv->rss) { - ret = rxq_promiscuous_enable(&priv->rxq_parent); + ret = rxq_promiscuous_enable(LIST_FIRST(&priv->parents)); if (ret) { priv_unlock(priv); return; @@ -4726,12 +4807,12 @@ mlx4_promiscuous_disable(struct rte_eth_dev *dev) if (mlx4_is_secondary()) return; priv_lock(priv); - if (!priv->promisc) { + if (!priv->promisc || priv->isolated) { priv_unlock(priv); return; } if (priv->rss) { - rxq_promiscuous_disable(&priv->rxq_parent); + rxq_promiscuous_disable(LIST_FIRST(&priv->parents)); goto end; } for (i = 0; (i != priv->rxqs_n); ++i) @@ -4758,6 +4839,12 @@ mlx4_allmulticast_enable(struct rte_eth_dev *dev) if (mlx4_is_secondary()) return; priv_lock(priv); + if (priv->isolated) { + DEBUG("%p: cannot enable allmulticast, " + "device is in isolated mode", (void *)dev); + priv_unlock(priv); + return; + } if (priv->allmulti) { priv_unlock(priv); return; @@ -4766,7 +4853,7 @@ mlx4_allmulticast_enable(struct rte_eth_dev *dev) if (!priv->started) goto end; if (priv->rss) { - ret = rxq_allmulticast_enable(&priv->rxq_parent); + ret = rxq_allmulticast_enable(LIST_FIRST(&priv->parents)); if (ret) { priv_unlock(priv); return; @@ -4806,12 +4893,12 @@ mlx4_allmulticast_disable(struct rte_eth_dev *dev) if (mlx4_is_secondary()) return; priv_lock(priv); - if (!priv->allmulti) { + if (!priv->allmulti || priv->isolated) { priv_unlock(priv); return; } if (priv->rss) { - rxq_allmulticast_disable(&priv->rxq_parent); + rxq_allmulticast_disable(LIST_FIRST(&priv->parents)); goto end; } for (i = 0; (i != priv->rxqs_n); ++i) @@ -4877,6 +4964,10 @@ mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete) return -1; } +static int +mlx4_ibv_device_to_pci_addr(const struct ibv_device *device, + struct rte_pci_addr *pci_addr); + /** * DPDK callback to change the MTU. * @@ -4925,21 +5016,16 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) /* Reconfigure each RX queue. */ for (i = 0; (i != priv->rxqs_n); ++i) { struct rxq *rxq = (*priv->rxqs)[i]; - unsigned int mb_len; unsigned int max_frame_len; - int sp; if (rxq == NULL) continue; - /* Calculate new maximum frame length according to MTU and - * toggle scattered support (sp) if necessary. */ + /* Calculate new maximum frame length according to MTU. */ max_frame_len = (priv->mtu + ETHER_HDR_LEN + (ETHER_MAX_VLAN_FRAME_LEN - ETHER_MAX_LEN)); - mb_len = rte_pktmbuf_data_room_size(rxq->mp); - assert(mb_len >= RTE_PKTMBUF_HEADROOM); - sp = (max_frame_len > (mb_len - RTE_PKTMBUF_HEADROOM)); /* Provide new values to rxq_setup(). */ - dev->data->dev_conf.rxmode.jumbo_frame = sp; + dev->data->dev_conf.rxmode.jumbo_frame = + (max_frame_len > ETHER_MAX_LEN); dev->data->dev_conf.rxmode.max_rx_pkt_len = max_frame_len; ret = rxq_rehash(dev, rxq); if (ret) { @@ -4950,7 +5036,7 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) } /* Reenable non-RSS queue attributes. No need to check * for errors at this stage. */ - if (!priv->rss) { + if (!priv->rss && !priv->isolated) { rxq_mac_addrs_add(rxq); if (priv->promisc) rxq_promiscuous_enable(rxq); @@ -5121,7 +5207,7 @@ vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) * Rehashing flows in all RX queues is necessary. */ if (priv->rss) - rxq_mac_addrs_del(&priv->rxq_parent); + rxq_mac_addrs_del(LIST_FIRST(&priv->parents)); else for (i = 0; (i != priv->rxqs_n); ++i) if ((*priv->rxqs)[i] != NULL) @@ -5129,7 +5215,7 @@ vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) priv->vlan_filter[j].enabled = 1; if (priv->started) { if (priv->rss) - rxq_mac_addrs_add(&priv->rxq_parent); + rxq_mac_addrs_add(LIST_FIRST(&priv->parents)); else for (i = 0; (i != priv->rxqs_n); ++i) { if ((*priv->rxqs)[i] == NULL) @@ -5143,7 +5229,7 @@ vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) * Rehashing flows in all RX queues is necessary. */ if (priv->rss) - rxq_mac_addrs_del(&priv->rxq_parent); + rxq_mac_addrs_del(LIST_FIRST(&priv->parents)); else for (i = 0; (i != priv->rxqs_n); ++i) if ((*priv->rxqs)[i] != NULL) @@ -5151,7 +5237,7 @@ vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) priv->vlan_filter[j].enabled = 0; if (priv->started) { if (priv->rss) - rxq_mac_addrs_add(&priv->rxq_parent); + rxq_mac_addrs_add(LIST_FIRST(&priv->parents)); else for (i = 0; (i != priv->rxqs_n); ++i) { if ((*priv->rxqs)[i] == NULL) @@ -5185,12 +5271,68 @@ mlx4_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) if (mlx4_is_secondary()) return -E_RTE_SECONDARY; priv_lock(priv); + if (priv->isolated) { + DEBUG("%p: cannot set vlan filter, " + "device is in isolated mode", (void *)dev); + priv_unlock(priv); + return -EINVAL; + } ret = vlan_filter_set(dev, vlan_id, on); priv_unlock(priv); assert(ret >= 0); return -ret; } +const struct rte_flow_ops mlx4_flow_ops = { + .validate = mlx4_flow_validate, + .create = mlx4_flow_create, + .destroy = mlx4_flow_destroy, + .flush = mlx4_flow_flush, + .query = NULL, + .isolate = mlx4_flow_isolate, +}; + +/** + * Manage filter operations. + * + * @param dev + * Pointer to Ethernet device structure. + * @param filter_type + * Filter type. + * @param filter_op + * Operation to perform. + * @param arg + * Pointer to operation-specific structure. + * + * @return + * 0 on success, negative errno value on failure. + */ +static int +mlx4_dev_filter_ctrl(struct rte_eth_dev *dev, + enum rte_filter_type filter_type, + enum rte_filter_op filter_op, + void *arg) +{ + int ret = EINVAL; + + switch (filter_type) { + case RTE_ETH_FILTER_GENERIC: + if (filter_op != RTE_ETH_FILTER_GET) + return -EINVAL; + *(const void **)arg = &mlx4_flow_ops; + return 0; + case RTE_ETH_FILTER_FDIR: + DEBUG("%p: filter type FDIR is not supported by this PMD", + (void *)dev); + break; + default: + ERROR("%p: filter type (%d) not supported", + (void *)dev, filter_type); + break; + } + return -ret; +} + static const struct eth_dev_ops mlx4_dev_ops = { .dev_configure = mlx4_dev_configure, .dev_start = mlx4_dev_start, @@ -5225,6 +5367,9 @@ static const struct eth_dev_ops mlx4_dev_ops = { .mac_addr_add = mlx4_mac_addr_add, .mac_addr_set = mlx4_mac_addr_set, .mtu_set = mlx4_dev_set_mtu, + .filter_ctrl = mlx4_dev_filter_ctrl, + .rx_queue_intr_enable = mlx4_rx_intr_enable, + .rx_queue_intr_disable = mlx4_rx_intr_disable, }; /** @@ -5264,7 +5409,7 @@ mlx4_ibv_device_to_pci_addr(const struct ibv_device *device, /* Extract information. */ if (sscanf(line, "PCI_SLOT_NAME=" - "%" SCNx16 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n", + "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n", &pci_addr->domain, &pci_addr->bus, &pci_addr->devid, @@ -5355,54 +5500,63 @@ mlx4_getenv_int(const char *name) static void mlx4_dev_link_status_handler(void *); static void -mlx4_dev_interrupt_handler(struct rte_intr_handle *, void *); +mlx4_dev_interrupt_handler(void *); /** - * Link status handler. + * Link/device status handler. * * @param priv * Pointer to private structure. * @param dev * Pointer to the rte_eth_dev structure. + * @param events + * Pointer to event flags holder. * * @return - * Nonzero if the callback process can be called immediately. + * Number of events */ static int -priv_dev_link_status_handler(struct priv *priv, struct rte_eth_dev *dev) +priv_dev_status_handler(struct priv *priv, struct rte_eth_dev *dev, + uint32_t *events) { struct ibv_async_event event; int port_change = 0; + struct rte_eth_link *link = &dev->data->dev_link; int ret = 0; + *events = 0; /* Read all message and acknowledge them. */ for (;;) { if (ibv_get_async_event(priv->ctx, &event)) break; - - if (event.event_type == IBV_EVENT_PORT_ACTIVE || - event.event_type == IBV_EVENT_PORT_ERR) + if ((event.event_type == IBV_EVENT_PORT_ACTIVE || + event.event_type == IBV_EVENT_PORT_ERR) && + (priv->intr_conf.lsc == 1)) { port_change = 1; - else + ret++; + } else if (event.event_type == IBV_EVENT_DEVICE_FATAL && + priv->intr_conf.rmv == 1) { + *events |= (1 << RTE_ETH_EVENT_INTR_RMV); + ret++; + } else DEBUG("event type %d on port %d not handled", event.event_type, event.element.port_num); ibv_ack_async_event(&event); } - - if (port_change ^ priv->pending_alarm) { - struct rte_eth_link *link = &dev->data->dev_link; - - priv->pending_alarm = 0; - mlx4_link_update(dev, 0); - if (((link->link_speed == 0) && link->link_status) || - ((link->link_speed != 0) && !link->link_status)) { + if (!port_change) + return ret; + mlx4_link_update(dev, 0); + if (((link->link_speed == 0) && link->link_status) || + ((link->link_speed != 0) && !link->link_status)) { + if (!priv->pending_alarm) { /* Inconsistent status, check again later. */ priv->pending_alarm = 1; rte_eal_alarm_set(MLX4_ALARM_TIMEOUT_US, mlx4_dev_link_status_handler, dev); - } else - ret = 1; + } + } else { + *events |= (1 << RTE_ETH_EVENT_INTR_LSC); } return ret; } @@ -5418,14 +5572,17 @@ mlx4_dev_link_status_handler(void *arg) { struct rte_eth_dev *dev = arg; struct priv *priv = dev->data->dev_private; + uint32_t events; int ret; priv_lock(priv); assert(priv->pending_alarm == 1); - ret = priv_dev_link_status_handler(priv, dev); + priv->pending_alarm = 0; + ret = priv_dev_status_handler(priv, dev, &events); priv_unlock(priv); - if (ret) - _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL); + if (ret > 0 && events & (1 << RTE_ETH_EVENT_INTR_LSC)) + _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL, + NULL); } /** @@ -5437,18 +5594,32 @@ mlx4_dev_link_status_handler(void *arg) * Callback argument. */ static void -mlx4_dev_interrupt_handler(struct rte_intr_handle *intr_handle, void *cb_arg) +mlx4_dev_interrupt_handler(void *cb_arg) { struct rte_eth_dev *dev = cb_arg; struct priv *priv = dev->data->dev_private; int ret; + uint32_t ev; + int i; - (void)intr_handle; priv_lock(priv); - ret = priv_dev_link_status_handler(priv, dev); + ret = priv_dev_status_handler(priv, dev, &ev); priv_unlock(priv); - if (ret) - _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL); + if (ret > 0) { + for (i = RTE_ETH_EVENT_UNKNOWN; + i < RTE_ETH_EVENT_MAX; + i++) { + if (ev & (1 << i)) { + ev &= ~(1 << i); + _rte_eth_dev_callback_process(dev, i, NULL, + NULL); + ret--; + } + } + if (ret) + WARN("%d event%s not processed", ret, + (ret > 1 ? "s were" : " was")); + } } /** @@ -5458,20 +5629,30 @@ mlx4_dev_interrupt_handler(struct rte_intr_handle *intr_handle, void *cb_arg) * Pointer to private structure. * @param dev * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative errno value on failure. */ -static void +static int priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev) { - if (!dev->data->dev_conf.intr_conf.lsc) - return; - rte_intr_callback_unregister(&priv->intr_handle, - mlx4_dev_interrupt_handler, - dev); - if (priv->pending_alarm) - rte_eal_alarm_cancel(mlx4_dev_link_status_handler, dev); - priv->pending_alarm = 0; + int ret; + + if (priv->intr_conf.lsc || + priv->intr_conf.rmv) + return 0; + ret = rte_intr_callback_unregister(&priv->intr_handle, + mlx4_dev_interrupt_handler, + dev); + if (ret < 0) { + ERROR("rte_intr_callback_unregister failed with %d" + "%s%s%s", ret, + (errno ? " (errno: " : ""), + (errno ? strerror(errno) : ""), + (errno ? ")" : "")); + } priv->intr_handle.fd = 0; priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; + return ret; } /** @@ -5481,30 +5662,377 @@ priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev) * Pointer to private structure. * @param dev * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative errno value on failure. */ -static void -priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev) +static int +priv_dev_interrupt_handler_install(struct priv *priv, + struct rte_eth_dev *dev) { - int rc, flags; + int flags; + int rc; - if (!dev->data->dev_conf.intr_conf.lsc) - return; + /* Check whether the interrupt handler has already been installed + * for either type of interrupt + */ + if (priv->intr_conf.lsc && + priv->intr_conf.rmv && + priv->intr_handle.fd) + return 0; assert(priv->ctx->async_fd > 0); flags = fcntl(priv->ctx->async_fd, F_GETFL); rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK); if (rc < 0) { INFO("failed to change file descriptor async event queue"); dev->data->dev_conf.intr_conf.lsc = 0; + dev->data->dev_conf.intr_conf.rmv = 0; + return -errno; } else { priv->intr_handle.fd = priv->ctx->async_fd; priv->intr_handle.type = RTE_INTR_HANDLE_EXT; - rte_intr_callback_register(&priv->intr_handle, - mlx4_dev_interrupt_handler, - dev); + rc = rte_intr_callback_register(&priv->intr_handle, + mlx4_dev_interrupt_handler, + dev); + if (rc) { + ERROR("rte_intr_callback_register failed " + " (errno: %s)", strerror(errno)); + return rc; + } + } + return 0; +} + +/** + * Uninstall interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative value on error. + */ +static int +priv_dev_removal_interrupt_handler_uninstall(struct priv *priv, + struct rte_eth_dev *dev) +{ + if (dev->data->dev_conf.intr_conf.rmv) { + priv->intr_conf.rmv = 0; + return priv_dev_interrupt_handler_uninstall(priv, dev); + } + return 0; +} + +/** + * Uninstall interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative value on error, + */ +static int +priv_dev_link_interrupt_handler_uninstall(struct priv *priv, + struct rte_eth_dev *dev) +{ + int ret = 0; + + if (dev->data->dev_conf.intr_conf.lsc) { + priv->intr_conf.lsc = 0; + ret = priv_dev_interrupt_handler_uninstall(priv, dev); + if (ret) + return ret; + } + if (priv->pending_alarm) + if (rte_eal_alarm_cancel(mlx4_dev_link_status_handler, + dev)) { + ERROR("rte_eal_alarm_cancel failed " + " (errno: %s)", strerror(rte_errno)); + return -rte_errno; + } + priv->pending_alarm = 0; + return 0; +} + +/** + * Install link interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative value on error. + */ +static int +priv_dev_link_interrupt_handler_install(struct priv *priv, + struct rte_eth_dev *dev) +{ + int ret; + + if (dev->data->dev_conf.intr_conf.lsc) { + ret = priv_dev_interrupt_handler_install(priv, dev); + if (ret) + return ret; + priv->intr_conf.lsc = 1; + } + return 0; +} + +/** + * Install removal interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative value on error. + */ +static int +priv_dev_removal_interrupt_handler_install(struct priv *priv, + struct rte_eth_dev *dev) +{ + int ret; + + if (dev->data->dev_conf.intr_conf.rmv) { + ret = priv_dev_interrupt_handler_install(priv, dev); + if (ret) + return ret; + priv->intr_conf.rmv = 1; + } + return 0; +} + +/** + * Allocate queue vector and fill epoll fd list for Rx interrupts. + * + * @param priv + * Pointer to private structure. + * + * @return + * 0 on success, negative on failure. + */ +static int +priv_rx_intr_vec_enable(struct priv *priv) +{ + unsigned int i; + unsigned int rxqs_n = priv->rxqs_n; + unsigned int n = RTE_MIN(rxqs_n, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID); + unsigned int count = 0; + struct rte_intr_handle *intr_handle = priv->dev->intr_handle; + + if (!priv->dev->data->dev_conf.intr_conf.rxq) + return 0; + priv_rx_intr_vec_disable(priv); + intr_handle->intr_vec = malloc(sizeof(intr_handle->intr_vec[rxqs_n])); + if (intr_handle->intr_vec == NULL) { + ERROR("failed to allocate memory for interrupt vector," + " Rx interrupts will not be supported"); + return -ENOMEM; + } + intr_handle->type = RTE_INTR_HANDLE_EXT; + for (i = 0; i != n; ++i) { + struct rxq *rxq = (*priv->rxqs)[i]; + int fd; + int flags; + int rc; + + /* Skip queues that cannot request interrupts. */ + if (!rxq || !rxq->channel) { + /* Use invalid intr_vec[] index to disable entry. */ + intr_handle->intr_vec[i] = + RTE_INTR_VEC_RXTX_OFFSET + + RTE_MAX_RXTX_INTR_VEC_ID; + continue; + } + if (count >= RTE_MAX_RXTX_INTR_VEC_ID) { + ERROR("too many Rx queues for interrupt vector size" + " (%d), Rx interrupts cannot be enabled", + RTE_MAX_RXTX_INTR_VEC_ID); + priv_rx_intr_vec_disable(priv); + return -1; + } + fd = rxq->channel->fd; + flags = fcntl(fd, F_GETFL); + rc = fcntl(fd, F_SETFL, flags | O_NONBLOCK); + if (rc < 0) { + ERROR("failed to make Rx interrupt file descriptor" + " %d non-blocking for queue index %d", fd, i); + priv_rx_intr_vec_disable(priv); + return rc; + } + intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + count; + intr_handle->efds[count] = fd; + count++; + } + if (!count) + priv_rx_intr_vec_disable(priv); + else + intr_handle->nb_efd = count; + return 0; +} + +/** + * Clean up Rx interrupts handler. + * + * @param priv + * Pointer to private structure. + */ +static void +priv_rx_intr_vec_disable(struct priv *priv) +{ + struct rte_intr_handle *intr_handle = priv->dev->intr_handle; + + rte_intr_free_epoll_fd(intr_handle); + free(intr_handle->intr_vec); + intr_handle->nb_efd = 0; + intr_handle->intr_vec = NULL; +} + +/** + * DPDK callback for Rx queue interrupt enable. + * + * @param dev + * Pointer to Ethernet device structure. + * @param idx + * Rx queue index. + * + * @return + * 0 on success, negative on failure. + */ +static int +mlx4_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx) +{ + struct priv *priv = dev->data->dev_private; + struct rxq *rxq = (*priv->rxqs)[idx]; + int ret; + + if (!rxq || !rxq->channel) + ret = EINVAL; + else + ret = ibv_req_notify_cq(rxq->cq, 0); + if (ret) + WARN("unable to arm interrupt on rx queue %d", idx); + return -ret; +} + +/** + * DPDK callback for Rx queue interrupt disable. + * + * @param dev + * Pointer to Ethernet device structure. + * @param idx + * Rx queue index. + * + * @return + * 0 on success, negative on failure. + */ +static int +mlx4_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx) +{ + struct priv *priv = dev->data->dev_private; + struct rxq *rxq = (*priv->rxqs)[idx]; + struct ibv_cq *ev_cq; + void *ev_ctx; + int ret; + + if (!rxq || !rxq->channel) { + ret = EINVAL; + } else { + ret = ibv_get_cq_event(rxq->cq->channel, &ev_cq, &ev_ctx); + if (ret || ev_cq != rxq->cq) + ret = EINVAL; } + if (ret) + WARN("unable to disable interrupt on rx queue %d", + idx); + else + ibv_ack_cq_events(rxq->cq, 1); + return -ret; } -static struct eth_driver mlx4_driver; +/** + * Verify and store value for device argument. + * + * @param[in] key + * Key argument to verify. + * @param[in] val + * Value associated with key. + * @param out + * User data. + * + * @return + * 0 on success, negative errno value on failure. + */ +static int +mlx4_arg_parse(const char *key, const char *val, void *out) +{ + struct mlx4_conf *conf = out; + unsigned long tmp; + + errno = 0; + tmp = strtoul(val, NULL, 0); + if (errno) { + WARN("%s: \"%s\" is not a valid integer", key, val); + return -errno; + } + if (strcmp(MLX4_PMD_PORT_KVARG, key) == 0) { + if (tmp >= MLX4_PMD_MAX_PHYS_PORTS) { + ERROR("invalid port index %lu (max: %u)", + tmp, MLX4_PMD_MAX_PHYS_PORTS - 1); + return -EINVAL; + } + conf->active_ports |= 1 << tmp; + } else { + WARN("%s: unknown parameter", key); + return -EINVAL; + } + return 0; +} + +/** + * Parse device parameters. + * + * @param devargs + * Device arguments structure. + * + * @return + * 0 on success, negative errno value on failure. + */ +static int +mlx4_args(struct rte_devargs *devargs, struct mlx4_conf *conf) +{ + struct rte_kvargs *kvlist; + unsigned int arg_count; + int ret = 0; + int i; + + if (devargs == NULL) + return 0; + kvlist = rte_kvargs_parse(devargs->args, pmd_mlx4_init_params); + if (kvlist == NULL) { + ERROR("failed to parse kvargs"); + return -EINVAL; + } + /* Process parameters. */ + for (i = 0; pmd_mlx4_init_params[i]; ++i) { + arg_count = rte_kvargs_count(kvlist, MLX4_PMD_PORT_KVARG); + while (arg_count-- > 0) { + ret = rte_kvargs_process(kvlist, MLX4_PMD_PORT_KVARG, + mlx4_arg_parse, conf); + if (ret != 0) + goto free_kvlist; + } + } +free_kvlist: + rte_kvargs_free(kvlist); + return ret; +} + +static struct rte_pci_driver mlx4_driver; /** * DPDK callback to register a PCI device. @@ -5528,12 +6056,15 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) int err = 0; struct ibv_context *attr_ctx = NULL; struct ibv_device_attr device_attr; + struct mlx4_conf conf = { + .active_ports = 0, + }; unsigned int vf; int idx; int i; (void)pci_drv; - assert(pci_drv == &mlx4_driver.pci_drv); + assert(pci_drv == &mlx4_driver); /* Get mlx4_dev[] index. */ idx = mlx4_dev_idx(&pci_dev->addr); if (idx == -1) { @@ -5547,10 +6078,8 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) list = ibv_get_device_list(&i); if (list == NULL) { assert(errno); - if (errno == ENOSYS) { - WARN("cannot list devices, is ib_uverbs loaded?"); - return 0; - } + if (errno == ENOSYS) + ERROR("cannot list devices, is ib_uverbs loaded?"); return -errno; } assert(i >= 0); @@ -5582,11 +6111,11 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) ibv_free_device_list(list); switch (err) { case 0: - WARN("cannot access device, is mlx4_ib loaded?"); - return 0; + ERROR("cannot access device, is mlx4_ib loaded?"); + return -ENODEV; case EINVAL: - WARN("cannot use device, are drivers up to date?"); - return 0; + ERROR("cannot use device, are drivers up to date?"); + return -EINVAL; } assert(err > 0); return -err; @@ -5598,6 +6127,15 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) goto error; INFO("%u port(s) detected", device_attr.phys_port_cnt); + if (mlx4_args(pci_dev->device.devargs, &conf)) { + ERROR("failed to process device arguments"); + goto error; + } + /* Use all ports when none are defined */ + if (conf.active_ports == 0) { + for (i = 0; i < MLX4_PMD_MAX_PHYS_PORTS; i++) + conf.active_ports |= 1 << i; + } for (i = 0; i < device_attr.phys_port_cnt; i++) { uint32_t port = i + 1; /* ports are indexed from one */ uint32_t test = (1 << i); @@ -5611,6 +6149,9 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) #endif /* HAVE_EXP_QUERY_DEVICE */ struct ether_addr mac; + /* If port is not active, skip. */ + if (!(conf.active_ports & (1 << i))) + continue; #ifdef HAVE_EXP_QUERY_DEVICE exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS; #ifdef RSS_SUPPORT @@ -5822,7 +6363,16 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) rte_eth_copy_pci_info(eth_dev, pci_dev); - eth_dev->driver = &mlx4_driver; + eth_dev->device->driver = &mlx4_driver.driver; + + /* + * Copy and override interrupt handle to prevent it from + * being shared between all ethdev instances of a given PCI + * device. This is required to properly handle Rx interrupts + * on all ports. + */ + priv->intr_handle_dev = *eth_dev->intr_handle; + eth_dev->intr_handle = &priv->intr_handle_dev; priv->dev = eth_dev; eth_dev->dev_ops = &mlx4_dev_ops; @@ -5830,6 +6380,9 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) /* Bring Ethernet device up. */ DEBUG("forcing Ethernet interface up"); priv_set_flags(priv, ~IFF_UP, IFF_UP); + /* Update link status once if waiting for LSC. */ + if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) + mlx4_link_update(eth_dev, 0); continue; port_error: @@ -5883,16 +6436,14 @@ static const struct rte_pci_id mlx4_pci_id_map[] = { } }; -static struct eth_driver mlx4_driver = { - .pci_drv = { - .driver = { - .name = MLX4_DRIVER_NAME - }, - .id_table = mlx4_pci_id_map, - .probe = mlx4_pci_probe, - .drv_flags = RTE_PCI_DRV_INTR_LSC, +static struct rte_pci_driver mlx4_driver = { + .driver = { + .name = MLX4_DRIVER_NAME }, - .dev_private_size = sizeof(struct priv) + .id_table = mlx4_pci_id_map, + .probe = mlx4_pci_probe, + .drv_flags = RTE_PCI_DRV_INTR_LSC | + RTE_PCI_DRV_INTR_RMV, }; /** @@ -5911,7 +6462,7 @@ rte_mlx4_pmd_init(void) */ setenv("RDMAV_HUGEPAGES_SAFE", "1", 1); ibv_fork_init(); - rte_eal_pci_register(&mlx4_driver.pci_drv); + rte_pci_register(&mlx4_driver); } RTE_PMD_EXPORT_NAME(net_mlx4, __COUNTER__);