X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx4%2Fmlx4.c;h=394b87c4c21e7799ef8d699d910f19d04d10b980;hb=586db080585b;hp=6c6342f644719d8bcd3423d42cd98f577c556b05;hpb=ab351fe1c95cc56f3c73066b1697f5590ef5c301;p=dpdk.git diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c index 6c6342f644..394b87c4c2 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 6WIND S.A. + * Copyright 2012 Mellanox * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -34,7 +34,6 @@ /* * Known limitations: * - RSS hash key and options cannot be modified. - * - Hardware counters aren't implemented. */ /* System headers. */ @@ -54,27 +53,13 @@ #include #include #include -#include #include #include +#include -/* Verbs header. */ -/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */ -#ifdef PEDANTIC -#pragma GCC diagnostic ignored "-pedantic" -#endif -#include -#ifdef PEDANTIC -#pragma GCC diagnostic error "-pedantic" -#endif - -/* DPDK headers don't like -pedantic. */ -#ifdef PEDANTIC -#pragma GCC diagnostic ignored "-pedantic" -#endif -#include #include #include +#include #include #include #include @@ -83,32 +68,20 @@ #include #include #include -#include #include -#ifdef PEDANTIC -#pragma GCC diagnostic error "-pedantic" -#endif +#include +#include +#include +#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) @@ -130,170 +103,37 @@ typedef union { #define WR_ID(o) (((wr_id_t *)&(o))->data) -/* Compile-time check. */ -static inline void wr_id_t_check(void) -{ - wr_id_t check[1 + (2 * -!(sizeof(wr_id_t) == sizeof(uint64_t)))]; - - (void)check; - (void)wr_id_t_check; -} - /* Transpose flags. Useful to convert IBV to DPDK flags. */ #define TRANSPOSE(val, from, to) \ (((from) >= (to)) ? \ (((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. */ +/** Configuration structure for device arguments. */ +struct mlx4_conf { + struct { + uint32_t present; /**< Bit-field for existing ports. */ + uint32_t enabled; /**< Bit-field for user-enabled ports. */ + } ports; }; -/* 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. */ +/* Available parameters list. */ +const char *pmd_mlx4_init_params[] = { + MLX4_PMD_PORT_KVARG, + NULL, }; -/* 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. */ - uint32_t mb_len; /* Length of a mp-issued mbuf. */ - struct mlx4_rxq_stats stats; /* RX queue counters. */ - unsigned int socket; /* CPU socket ID for allocations. */ - struct ibv_exp_res_domain *rd; /* Resource Domain. */ -}; +static int +mlx4_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx); -/* TX element. */ -struct txq_elt { - struct rte_mbuf *buf; -}; +static int +mlx4_rx_intr_disable(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 { - 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 +priv_rx_intr_vec_enable(struct priv *priv); -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. */ -#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. */ - rte_spinlock_t lock; /* Lock for control functions. */ -}; +static void +priv_rx_intr_vec_disable(struct priv *priv); /** * Lock private structure to protect it from concurrent access in the @@ -302,8 +142,7 @@ struct priv { * @param priv * Pointer to private structure. */ -static void -priv_lock(struct priv *priv) +void priv_lock(struct priv *priv) { rte_spinlock_lock(&priv->lock); } @@ -314,8 +153,7 @@ 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); } @@ -624,7 +462,15 @@ priv_get_mtu(struct priv *priv, uint16_t *mtu) static int priv_set_mtu(struct priv *priv, uint16_t mtu) { - return priv_set_sysfs_ulong(priv, "mtu", mtu); + uint16_t new_mtu; + + if (priv_set_sysfs_ulong(priv, "mtu", mtu) || + priv_get_mtu(priv, &new_mtu)) + return -1; + if (new_mtu == mtu) + return 0; + errno = EINVAL; + return -1; } /** @@ -648,20 +494,107 @@ priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags) if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1) return -1; tmp &= keep; - tmp |= flags; + tmp |= (flags & (~keep)); return priv_set_sysfs_ulong(priv, "flags", tmp); } /* Device configuration. */ +static int +txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, + unsigned int socket, const struct rte_eth_txconf *conf); + +static void +txq_cleanup(struct txq *txq); + static int rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, - unsigned int socket, 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. * @@ -681,7 +614,6 @@ 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; @@ -692,6 +624,15 @@ dev_configure(struct rte_eth_dev *dev) } if (rxqs_n == priv->rxqs_n) return 0; + if (!rte_is_power_of_2(rxqs_n) && !priv->isolated) { + unsigned n_active; + + n_active = rte_align32pow2(rxqs_n + 1) >> 1; + WARN("%p: number of RX queues must be a power" + " of 2: %u queues among %u will be active", + (void *)dev, n_active, rxqs_n); + } + INFO("%p: RX queues number update: %u -> %u", (void *)dev, priv->rxqs_n, rxqs_n); /* If RSS is enabled, disable it first. */ @@ -702,7 +643,7 @@ 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; } @@ -727,14 +668,14 @@ 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, 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; } /** @@ -759,6 +700,9 @@ mlx4_dev_configure(struct rte_eth_dev *dev) return -ret; } +static uint16_t mlx4_tx_burst(void *, struct rte_mbuf **, uint16_t); +static uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t); + /* TX queues handling. */ /** @@ -791,7 +735,7 @@ txq_alloc_elts(struct txq *txq, unsigned int elts_n) } mr_linear = ibv_reg_mr(txq->priv->pd, elts_linear, sizeof(*elts_linear), - (IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE)); + IBV_ACCESS_LOCAL_WRITE); if (mr_linear == NULL) { ERROR("%p: unable to configure MR, ibv_reg_mr() failed", (void *)txq); @@ -840,14 +784,20 @@ error: static void txq_free_elts(struct txq *txq) { - unsigned int i; unsigned int elts_n = txq->elts_n; + unsigned int elts_head = txq->elts_head; + unsigned int elts_tail = txq->elts_tail; struct txq_elt (*elts)[elts_n] = txq->elts; linear_t (*elts_linear)[elts_n] = txq->elts_linear; struct ibv_mr *mr_linear = txq->mr_linear; DEBUG("%p: freeing WRs", (void *)txq); txq->elts_n = 0; + txq->elts_head = 0; + txq->elts_tail = 0; + txq->elts_comp = 0; + txq->elts_comp_cd = 0; + txq->elts_comp_cd_init = 0; txq->elts = NULL; txq->elts_linear = NULL; txq->mr_linear = NULL; @@ -857,12 +807,17 @@ txq_free_elts(struct txq *txq) rte_free(elts_linear); if (elts == NULL) return; - for (i = 0; (i != elemof(*elts)); ++i) { - struct txq_elt *elt = &(*elts)[i]; + while (elts_tail != elts_head) { + struct txq_elt *elt = &(*elts)[elts_tail]; - if (elt->buf == NULL) - continue; + assert(elt->buf != NULL); rte_pktmbuf_free(elt->buf); +#ifndef NDEBUG + /* Poisoning. */ + memset(elt, 0x77, sizeof(*elt)); +#endif + if (++elts_tail == elts_n) + elts_tail = 0; } rte_free(elts); } @@ -954,10 +909,6 @@ txq_complete(struct txq *txq) if (unlikely(elts_comp == 0)) return 0; -#ifdef DEBUG_SEND - DEBUG("%p: processing %u work requests completions", - (void *)txq, elts_comp); -#endif wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp); if (unlikely(wcs_n == 0)) return 0; @@ -980,6 +931,139 @@ txq_complete(struct txq *txq) return 0; } +struct mlx4_check_mempool_data { + int ret; + char *start; + char *end; +}; + +/* Called by mlx4_check_mempool() when iterating the memory chunks. */ +static void mlx4_check_mempool_cb(struct rte_mempool *mp, + void *opaque, struct rte_mempool_memhdr *memhdr, + unsigned mem_idx) +{ + struct mlx4_check_mempool_data *data = opaque; + + (void)mp; + (void)mem_idx; + + /* It already failed, skip the next chunks. */ + if (data->ret != 0) + return; + /* It is the first chunk. */ + if (data->start == NULL && data->end == NULL) { + data->start = memhdr->addr; + data->end = data->start + memhdr->len; + return; + } + if (data->end == memhdr->addr) { + data->end += memhdr->len; + return; + } + if (data->start == (char *)memhdr->addr + memhdr->len) { + data->start -= memhdr->len; + return; + } + /* Error, mempool is not virtually contigous. */ + data->ret = -1; +} + +/** + * Check if a mempool can be used: it must be virtually contiguous. + * + * @param[in] mp + * Pointer to memory pool. + * @param[out] start + * Pointer to the start address of the mempool virtual memory area + * @param[out] end + * Pointer to the end address of the mempool virtual memory area + * + * @return + * 0 on success (mempool is virtually contiguous), -1 on error. + */ +static int mlx4_check_mempool(struct rte_mempool *mp, uintptr_t *start, + uintptr_t *end) +{ + struct mlx4_check_mempool_data data; + + memset(&data, 0, sizeof(data)); + rte_mempool_mem_iter(mp, mlx4_check_mempool_cb, &data); + *start = (uintptr_t)data.start; + *end = (uintptr_t)data.end; + + return data.ret; +} + +/* For best performance, this function should not be inlined. */ +static struct ibv_mr *mlx4_mp2mr(struct ibv_pd *, struct rte_mempool *) + __rte_noinline; + +/** + * Register mempool as a memory region. + * + * @param pd + * Pointer to protection domain. + * @param mp + * Pointer to memory pool. + * + * @return + * Memory region pointer, NULL in case of error. + */ +static struct ibv_mr * +mlx4_mp2mr(struct ibv_pd *pd, struct rte_mempool *mp) +{ + const struct rte_memseg *ms = rte_eal_get_physmem_layout(); + uintptr_t start; + uintptr_t end; + unsigned int i; + + if (mlx4_check_mempool(mp, &start, &end) != 0) { + ERROR("mempool %p: not virtually contiguous", + (void *)mp); + return NULL; + } + + DEBUG("mempool %p area start=%p end=%p size=%zu", + (void *)mp, (void *)start, (void *)end, + (size_t)(end - start)); + /* Round start and end to page boundary if found in memory segments. */ + for (i = 0; (i < RTE_MAX_MEMSEG) && (ms[i].addr != NULL); ++i) { + uintptr_t addr = (uintptr_t)ms[i].addr; + size_t len = ms[i].len; + unsigned int align = ms[i].hugepage_sz; + + if ((start > addr) && (start < addr + len)) + start = RTE_ALIGN_FLOOR(start, align); + if ((end > addr) && (end < addr + len)) + end = RTE_ALIGN_CEIL(end, align); + } + DEBUG("mempool %p using start=%p end=%p size=%zu for MR", + (void *)mp, (void *)start, (void *)end, + (size_t)(end - start)); + return ibv_reg_mr(pd, + (void *)start, + end - start, + IBV_ACCESS_LOCAL_WRITE); +} + +/** + * Get Memory Pool (MP) from mbuf. If mbuf is indirect, the pool from which + * the cloned mbuf is allocated is returned instead. + * + * @param buf + * Pointer to mbuf. + * + * @return + * Memory pool where data is located for given mbuf. + */ +static struct rte_mempool * +txq_mb2mp(struct rte_mbuf *buf) +{ + if (unlikely(RTE_MBUF_INDIRECT(buf))) + return rte_mbuf_from_indirect(buf)->pool; + return buf->pool; +} + /** * Get Memory Region (MR) <-> Memory Pool (MP) association from txq->mp2mr[]. * Add MP to txq->mp2mr[] if it's not registered yet. If mp2mr[] is full, @@ -1011,11 +1095,9 @@ txq_mp2mr(struct txq *txq, struct rte_mempool *mp) } } /* Add a new entry, register MR first. */ - DEBUG("%p: discovered new memory pool %p", (void *)txq, (void *)mp); - mr = ibv_reg_mr(txq->priv->pd, - (void *)mp->elt_va_start, - (mp->elt_va_end - mp->elt_va_start), - (IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE)); + DEBUG("%p: discovered new memory pool \"%s\" (%p)", + (void *)txq, mp->name, (void *)mp); + mr = mlx4_mp2mr(txq->priv->pd, mp); if (unlikely(mr == NULL)) { DEBUG("%p: unable to configure MR, ibv_reg_mr() failed.", (void *)txq); @@ -1026,7 +1108,7 @@ txq_mp2mr(struct txq *txq, struct rte_mempool *mp) DEBUG("%p: MR <-> MP table full, dropping oldest entry.", (void *)txq); --i; - claim_zero(ibv_dereg_mr(txq->mp2mr[i].mr)); + claim_zero(ibv_dereg_mr(txq->mp2mr[0].mr)); memmove(&txq->mp2mr[0], &txq->mp2mr[1], (sizeof(txq->mp2mr) - sizeof(txq->mp2mr[0]))); } @@ -1034,12 +1116,65 @@ txq_mp2mr(struct txq *txq, struct rte_mempool *mp) txq->mp2mr[i].mp = mp; txq->mp2mr[i].mr = mr; txq->mp2mr[i].lkey = mr->lkey; - DEBUG("%p: new MR lkey for MP %p: 0x%08" PRIu32, - (void *)txq, (void *)mp, txq->mp2mr[i].lkey); + DEBUG("%p: new MR lkey for MP \"%s\" (%p): 0x%08" PRIu32, + (void *)txq, mp->name, (void *)mp, txq->mp2mr[i].lkey); return txq->mp2mr[i].lkey; } -#if MLX4_PMD_SGE_WR_N > 1 +struct txq_mp2mr_mbuf_check_data { + int ret; +}; + +/** + * Callback function for rte_mempool_obj_iter() to check whether a given + * mempool object looks like a mbuf. + * + * @param[in] mp + * The mempool pointer + * @param[in] arg + * Context data (struct txq_mp2mr_mbuf_check_data). Contains the + * return value. + * @param[in] obj + * Object address. + * @param index + * Object index, unused. + */ +static void +txq_mp2mr_mbuf_check(struct rte_mempool *mp, void *arg, void *obj, + uint32_t index __rte_unused) +{ + struct txq_mp2mr_mbuf_check_data *data = arg; + struct rte_mbuf *buf = obj; + + /* Check whether mbuf structure fits element size and whether mempool + * pointer is valid. */ + if (sizeof(*buf) > mp->elt_size || buf->pool != mp) + data->ret = -1; +} + +/** + * Iterator function for rte_mempool_walk() to register existing mempools and + * fill the MP to MR cache of a TX queue. + * + * @param[in] mp + * Memory Pool to register. + * @param *arg + * Pointer to TX queue structure. + */ +static void +txq_mp2mr_iter(struct rte_mempool *mp, void *arg) +{ + struct txq *txq = arg; + struct txq_mp2mr_mbuf_check_data data = { + .ret = 0, + }; + + /* Register mempool only if the first element looks like a mbuf. */ + if (rte_mempool_obj_iter(mp, txq_mp2mr_mbuf_check, &data) == 0 || + data.ret == -1) + return; + txq_mp2mr(txq, mp); +} /** * Copy scattered mbuf contents to a single linear buffer. @@ -1120,7 +1255,7 @@ tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt, uint32_t lkey; /* Retrieve Memory Region key for this memory pool. */ - lkey = txq_mp2mr(txq, buf->pool); + lkey = txq_mp2mr(txq, txq_mb2mp(buf)); if (unlikely(lkey == (uint32_t)-1)) { /* MR does not exist. */ DEBUG("%p: unable to get MP <-> MR association", @@ -1173,6 +1308,8 @@ tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt, sge->length = size; sge->lkey = txq->mr_linear->lkey; sent_size += size; + /* Include last segment. */ + segs++; } return (struct tx_burst_sg_ret){ .length = sent_size, @@ -1185,8 +1322,6 @@ stop: }; } -#endif /* MLX4_PMD_SGE_WR_N > 1 */ - /** * DPDK callback for TX. * @@ -1205,7 +1340,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) { struct txq *txq = (struct txq *)dpdk_txq; unsigned int elts_head = txq->elts_head; - const unsigned int elts_tail = txq->elts_tail; const unsigned int elts_n = txq->elts_n; unsigned int elts_comp_cd = txq->elts_comp_cd; unsigned int elts_comp = 0; @@ -1215,7 +1349,7 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) assert(elts_comp_cd != 0); txq_complete(txq); - max = (elts_n - (elts_head - elts_tail)); + max = (elts_n - (elts_head - txq->elts_tail)); if (max > elts_n) max -= elts_n; assert(max >= 1); @@ -1233,15 +1367,17 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) struct txq_elt *elt_next = &(*txq->elts)[elts_head_next]; struct txq_elt *elt = &(*txq->elts)[elts_head]; unsigned int segs = NB_SEGS(buf); -#ifdef MLX4_PMD_SOFT_COUNTERS unsigned int sent_size = 0; -#endif uint32_t send_flags = 0; /* Clean up old buffer. */ if (likely(elt->buf != NULL)) { struct rte_mbuf *tmp = elt->buf; +#ifndef NDEBUG + /* Poisoning. */ + memset(elt, 0x66, sizeof(*elt)); +#endif /* Faster than rte_pktmbuf_free(). */ do { struct rte_mbuf *next = NEXT(tmp); @@ -1276,7 +1412,7 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) addr = rte_pktmbuf_mtod(buf, uintptr_t); length = DATA_LEN(buf); /* Retrieve Memory Region key for this memory pool. */ - lkey = txq_mp2mr(txq, buf->pool); + lkey = txq_mp2mr(txq, txq_mb2mp(buf)); if (unlikely(lkey == (uint32_t)-1)) { /* MR does not exist. */ DEBUG("%p: unable to get MP <-> MR" @@ -1292,7 +1428,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) (uintptr_t)addr); RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf); /* Put packet into send queue. */ -#if MLX4_PMD_MAX_INLINE > 0 if (length <= txq->max_inline) err = txq->if_qp->send_pending_inline (txq->qp, @@ -1300,7 +1435,6 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) length, send_flags); else -#endif err = txq->if_qp->send_pending (txq->qp, addr, @@ -1309,11 +1443,8 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) send_flags); if (unlikely(err)) goto stop; -#ifdef MLX4_PMD_SOFT_COUNTERS sent_size += length; -#endif } else { -#if MLX4_PMD_SGE_WR_N > 1 struct ibv_sge sges[MLX4_PMD_SGE_WR_N]; struct tx_burst_sg_ret ret; @@ -1330,29 +1461,18 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) send_flags); if (unlikely(err)) goto stop; -#ifdef MLX4_PMD_SOFT_COUNTERS sent_size += ret.length; -#endif -#else /* MLX4_PMD_SGE_WR_N > 1 */ - DEBUG("%p: TX scattered buffers support not" - " compiled in", (void *)txq); - goto stop; -#endif /* MLX4_PMD_SGE_WR_N > 1 */ } elts_head = elts_head_next; -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increment sent bytes counter. */ txq->stats.obytes += sent_size; -#endif } stop: /* Take a shortcut if nothing must be sent. */ if (unlikely(i == 0)) return 0; -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increment sent packets counter. */ txq->stats.opackets += i; -#endif /* Ring QP doorbell. */ err = txq->if_qp->send_flush(txq->qp); if (unlikely(err)) { @@ -1404,6 +1524,8 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, int ret = 0; (void)conf; /* Thresholds configuration (ignored). */ + if (priv == NULL) + return EINVAL; if ((desc == 0) || (desc % MLX4_PMD_SGE_WR_N)) { ERROR("%p: invalid number of TX descriptors (must be a" " multiple of %d)", (void *)dev, MLX4_PMD_SGE_WR_N); @@ -1454,9 +1576,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, MLX4_PMD_SGE_WR_N) ? priv->device_attr.max_sge : MLX4_PMD_SGE_WR_N), -#if MLX4_PMD_MAX_INLINE > 0 .max_inline_data = MLX4_PMD_MAX_INLINE, -#endif }, .qp_type = IBV_QPT_RAW_PACKET, /* Do *NOT* enable this, completions events are managed per @@ -1474,10 +1594,8 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, (void *)dev, strerror(ret)); goto error; } -#if MLX4_PMD_MAX_INLINE > 0 /* ibv_create_qp() updates this value. */ tmpl.max_inline = attr.init.cap.max_inline_data; -#endif attr.mod = (struct ibv_exp_qp_attr){ /* Move the QP to this state. */ .qp_state = IBV_QPS_INIT, @@ -1547,6 +1665,8 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, txq_cleanup(txq); *txq = tmpl; DEBUG("%p: txq updated with %p", (void *)txq, (void *)&tmpl); + /* Pre-register known mempools. */ + rte_mempool_walk(txq_mp2mr_iter, txq); assert(ret == 0); return 0; error: @@ -2063,6 +2183,7 @@ 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 @@ -2206,11 +2327,12 @@ 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) @@ -2277,7 +2399,7 @@ 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; @@ -2455,14 +2577,17 @@ 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, @@ -2485,6 +2610,8 @@ rxq_cleanup(struct rxq *rxq) * @param flags * RX completion flags returned by poll_length_flags(). * + * @note: fix mlx4_dev_supported_ptypes_get() if any change here. + * * @return * Packet type for struct rte_mbuf. */ @@ -2496,19 +2623,25 @@ 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; } @@ -2530,25 +2663,20 @@ rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags) if (rxq->csum) ol_flags |= - TRANSPOSE(~flags, + TRANSPOSE(flags, IBV_EXP_CQ_RX_IP_CSUM_OK, - PKT_RX_IP_CKSUM_BAD) | - TRANSPOSE(~flags, + PKT_RX_IP_CKSUM_GOOD) | + TRANSPOSE(flags, IBV_EXP_CQ_RX_TCP_UDP_CSUM_OK, - PKT_RX_L4_CKSUM_BAD); - /* - * PKT_RX_IP_CKSUM_BAD and PKT_RX_L4_CKSUM_BAD are used in place - * of PKT_RX_EIP_CKSUM_BAD because the latter is not functional - * (its value is 0). - */ + PKT_RX_L4_CKSUM_GOOD); if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun)) ol_flags |= - TRANSPOSE(~flags, + TRANSPOSE(flags, IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK, - PKT_RX_IP_CKSUM_BAD) | - TRANSPOSE(~flags, + PKT_RX_IP_CKSUM_GOOD) | + TRANSPOSE(flags, IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK, - PKT_RX_L4_CKSUM_BAD); + PKT_RX_L4_CKSUM_GOOD); return ol_flags; } @@ -2631,10 +2759,8 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) " completion status (%d): %s", (void *)rxq, wc.wr_id, wc.status, ibv_wc_status_str(wc.status)); -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increment dropped packets counter. */ ++rxq->stats.idropped; -#endif /* Link completed WRs together for repost. */ *next = wr; next = &wr->next; @@ -2664,7 +2790,7 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) * cacheline while allocating rep. */ rte_prefetch0(seg); - rep = __rte_mbuf_raw_alloc(rxq->mp); + rep = rte_mbuf_raw_alloc(rxq->mp); if (unlikely(rep == NULL)) { /* * Unable to allocate a replacement mbuf, @@ -2691,9 +2817,15 @@ 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); - assert(rep->buf_len == rxq->mb_len); /* Reconfigure sge to use rep instead of seg. */ assert(sge->lkey == rxq->mr->lkey); sge->addr = ((uintptr_t)rep->buf_addr + seg_headroom); @@ -2735,14 +2867,13 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) PORT(pkt_buf) = rxq->port_id; PKT_LEN(pkt_buf) = pkt_buf_len; pkt_buf->packet_type = rxq_cq_to_pkt_type(flags); + pkt_buf->ol_flags = rxq_cq_to_ol_flags(rxq, flags); /* Return packet. */ *(pkts++) = pkt_buf; ++pkts_ret; -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increase bytes counter. */ rxq->stats.ibytes += pkt_buf_len; -#endif repost: if (++elts_head >= elts_n) elts_head = 0; @@ -2752,9 +2883,6 @@ repost: return 0; *next = NULL; /* Repost WRs. */ -#ifdef DEBUG_RECV - DEBUG("%p: reposting %d WRs", (void *)rxq, i); -#endif ret = ibv_post_recv(rxq->qp, head.next, &bad_wr); if (unlikely(ret)) { /* Inability to repost WRs is fatal. */ @@ -2765,10 +2893,8 @@ repost: abort(); } rxq->elts_head = elts_head; -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increase packets counter. */ rxq->stats.ipackets += pkts_ret; -#endif return pkts_ret; } @@ -2819,6 +2945,12 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) assert(wr->num_sge == 1); assert(elts_head < rxq->elts_n); assert(rxq->elts_head < rxq->elts_n); + /* + * Fetch initial bytes of packet descriptor into a + * cacheline while allocating rep. + */ + rte_mbuf_prefetch_part1(seg); + rte_mbuf_prefetch_part2(seg); ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL, &flags); if (unlikely(ret < 0)) { @@ -2843,10 +2975,8 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) " completion status (%d): %s", (void *)rxq, wc.wr_id, wc.status, ibv_wc_status_str(wc.status)); -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increment dropped packets counter. */ ++rxq->stats.idropped; -#endif /* Add SGE to array for repost. */ sges[i] = elt->sge; goto repost; @@ -2856,12 +2986,7 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) if (ret == 0) break; len = ret; - /* - * Fetch initial bytes of packet descriptor into a - * cacheline while allocating rep. - */ - rte_prefetch0(seg); - rep = __rte_mbuf_raw_alloc(rxq->mp); + rep = rte_mbuf_raw_alloc(rxq->mp); if (unlikely(rep == NULL)) { /* * Unable to allocate a replacement mbuf, @@ -2873,6 +2998,8 @@ 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; } @@ -2900,10 +3027,8 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) /* Return packet. */ *(pkts++) = seg; ++pkts_ret; -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increase bytes counter. */ rxq->stats.ibytes += len; -#endif repost: if (++elts_head >= elts_n) elts_head = 0; @@ -2912,9 +3037,6 @@ repost: if (unlikely(i == 0)) return 0; /* Repost WRs. */ -#ifdef DEBUG_RECV - DEBUG("%p: reposting %u WRs", (void *)rxq, i); -#endif ret = rxq->if_qp->recv_burst(rxq->qp, sges, i); if (unlikely(ret)) { /* Inability to repost WRs is fatal. */ @@ -2924,10 +3046,8 @@ repost: abort(); } rxq->elts_head = elts_head; -#ifdef MLX4_PMD_SOFT_COUNTERS /* Increase packets counter. */ rxq->stats.ipackets += pkts_ret; -#endif return pkts_ret; } @@ -2972,15 +3092,11 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc, .res_domain = rd, }; -#ifdef INLINE_RECV attr.max_inl_recv = priv->inl_recv_size; attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV; -#endif return ibv_exp_create_qp(priv->ctx, &attr); } -#ifdef RSS_SUPPORT - /** * Allocate a RSS Queue Pair. * Optionally setup inline receive if supported. @@ -2991,15 +3107,18 @@ 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. */ @@ -3025,26 +3144,23 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc, .res_domain = rd, }; -#ifdef INLINE_RECV 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 = priv->rxqs_n; + attr.qpg.parent_attrib.rss_child_count = + 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); } -#endif /* RSS_SUPPORT */ - /** * Reconfigure a RX queue with new parameters. * @@ -3071,14 +3187,10 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) unsigned int i, k; struct ibv_exp_qp_attr mod; 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. */ desc_n = (tmpl.elts_n * (tmpl.sp ? MLX4_PMD_SGE_WR_N : 1)); @@ -3093,9 +3205,10 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) rxq->csum_l2tun = tmpl.csum_l2tun; } /* Enable scattered packets support for this queue if necessary. */ - if ((dev->data->dev_conf.rxmode.jumbo_frame) && + assert(mb_len >= RTE_PKTMBUF_HEADROOM); + if (dev->data->dev_conf.rxmode.enable_scatter && (dev->data->dev_conf.rxmode.max_rx_pkt_len > - (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) { + (mb_len - RTE_PKTMBUF_HEADROOM))) { tmpl.sp = 1; desc_n /= MLX4_PMD_SGE_WR_N; } else @@ -3108,7 +3221,7 @@ 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); @@ -3121,6 +3234,8 @@ 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) { @@ -3128,12 +3243,6 @@ 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, @@ -3142,9 +3251,6 @@ 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", @@ -3152,8 +3258,15 @@ 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); @@ -3219,6 +3332,8 @@ 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 ? @@ -3246,46 +3361,160 @@ skip_rtr: } /** - * Configure a RX queue. + * Create verbs QP resources associated with a rxq. * - * @param dev - * Pointer to Ethernet device structure. * @param rxq * Pointer to RX queue structure. * @param desc * Number of descriptors to configure in queue. - * @param socket - * NUMA socket on which memory must be allocated. - * @param[in] conf - * Thresholds parameters. - * @param mp - * Memory pool for buffer allocations. + * @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. */ -static int -rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, - unsigned int socket, const struct rte_eth_rxconf *conf, - struct rte_mempool *mp) +int +rxq_create_qp(struct rxq *rxq, + uint16_t desc, + int inactive, + int children_n, + struct rxq *rxq_parent) { - struct priv *priv = dev->data->dev_private; - struct rxq tmpl = { - .priv = priv, - .mp = mp, - .socket = socket - }; + int ret; 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; + struct ibv_exp_query_intf_params params; enum ibv_exp_query_intf_status status; struct ibv_recv_wr *bad_wr; - struct rte_mbuf *buf; + int parent = (children_n > 0); + struct priv *priv = rxq->priv; + + if (priv->rss && !inactive && (rxq_parent || parent)) + rxq->qp = rxq_setup_qp_rss(priv, rxq->cq, desc, + children_n, rxq->rd, + rxq_parent); + else + 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 | + (parent ? IBV_EXP_QP_GROUP_RSS : 0) | + 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. + * + * @param dev + * Pointer to Ethernet device structure. + * @param rxq + * Pointer to RX queue structure. + * @param desc + * Number of descriptors to configure in queue. + * @param socket + * NUMA socket on which memory must be allocated. + * @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[in] conf + * 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, int children_n, + struct rxq *rxq_parent) +{ + struct priv *priv = dev->data->dev_private; + struct rxq tmpl = { + .priv = priv, + .mp = mp, + .socket = socket + }; + 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; + unsigned int mb_len; int ret = 0; - int parent = (rxq == &priv->rxq_parent); + int parent = (children_n > 0); (void)conf; /* Thresholds configuration (ignored). */ /* @@ -3299,42 +3528,37 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, desc = 1; goto skip_mr; } + mb_len = rte_pktmbuf_data_room_size(mp); if ((desc == 0) || (desc % MLX4_PMD_SGE_WR_N)) { ERROR("%p: invalid number of RX descriptors (must be a" " multiple of %d)", (void *)dev, MLX4_PMD_SGE_WR_N); return EINVAL; } - /* Get mbuf length. */ - buf = rte_pktmbuf_alloc(mp); - if (buf == NULL) { - ERROR("%p: unable to allocate mbuf", (void *)dev); - return ENOMEM; - } - tmpl.mb_len = buf->buf_len; - assert((rte_pktmbuf_headroom(buf) + - rte_pktmbuf_tailroom(buf)) == tmpl.mb_len); - assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM); - rte_pktmbuf_free(buf); /* Toggle RX checksum offload if hardware supports it. */ if (priv->hw_csum) tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum; if (priv->hw_csum_l2tun) tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum; /* Enable scattered packets support for this queue if necessary. */ - if ((dev->data->dev_conf.rxmode.jumbo_frame) && - (dev->data->dev_conf.rxmode.max_rx_pkt_len > - (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) { + assert(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); /* Use the entire RX mempool as the memory region. */ - tmpl.mr = ibv_reg_mr(priv->pd, - (void *)mp->elt_va_start, - (mp->elt_va_end - mp->elt_va_start), - (IBV_ACCESS_LOCAL_WRITE | - IBV_ACCESS_REMOTE_WRITE)); + tmpl.mr = mlx4_mp2mr(priv->pd, mp); if (tmpl.mr == NULL) { ret = EINVAL; ERROR("%p: MR creation failure: %s", @@ -3355,11 +3579,22 @@ 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", @@ -3370,45 +3605,6 @@ 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) - 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; @@ -3419,29 +3615,14 @@ 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; @@ -3453,21 +3634,11 @@ 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); @@ -3505,8 +3676,10 @@ 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; int ret; priv_lock(priv); @@ -3536,7 +3709,16 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, return -ENOMEM; } } - ret = rxq_setup(dev, rxq, desc, socket, 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 { @@ -3571,7 +3753,6 @@ 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", @@ -3584,6 +3765,15 @@ mlx4_rx_queue_release(void *dpdk_rxq) priv_unlock(priv); } +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. * @@ -3602,6 +3792,7 @@ mlx4_dev_start(struct rte_eth_dev *dev) unsigned int i = 0; unsigned int r; struct rxq *rxq; + int ret; priv_lock(priv); if (priv->started) { @@ -3610,8 +3801,11 @@ 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]; @@ -3619,8 +3813,6 @@ 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; @@ -3633,20 +3825,47 @@ 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; - return -ret; + goto err; } while ((--r) && ((rxq = (*priv->rxqs)[++i]), i)); + 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; } /** @@ -3672,13 +3891,17 @@ 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. */ @@ -3741,6 +3964,16 @@ removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) return 0; } +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. * @@ -3756,6 +3989,8 @@ mlx4_dev_close(struct rte_eth_dev *dev) void *tmp; unsigned int i; + if (priv == NULL) + return; priv_lock(priv); DEBUG("%p: closing device \"%s\"", (void *)dev, @@ -3794,17 +4029,104 @@ 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_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)); } +/** + * Change the link state (UP / DOWN). + * + * @param priv + * Pointer to Ethernet device private data. + * @param up + * Nonzero for link up, otherwise link down. + * + * @return + * 0 on success, errno value on failure. + */ +static int +priv_set_link(struct priv *priv, int up) +{ + struct rte_eth_dev *dev = priv->dev; + int err; + unsigned int i; + + if (up) { + err = priv_set_flags(priv, ~IFF_UP, IFF_UP); + if (err) + return err; + for (i = 0; i < priv->rxqs_n; i++) + if ((*priv->rxqs)[i]->sp) + break; + /* Check if an sp queue exists. + * Note: Some old frames might be received. + */ + if (i == priv->rxqs_n) + dev->rx_pkt_burst = mlx4_rx_burst; + else + dev->rx_pkt_burst = mlx4_rx_burst_sp; + dev->tx_pkt_burst = mlx4_tx_burst; + } else { + err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP); + if (err) + return err; + dev->rx_pkt_burst = removed_rx_burst; + dev->tx_pkt_burst = removed_tx_burst; + } + return 0; +} + +/** + * DPDK callback to bring the link DOWN. + * + * @param dev + * Pointer to Ethernet device structure. + * + * @return + * 0 on success, errno value on failure. + */ +static int +mlx4_set_link_down(struct rte_eth_dev *dev) +{ + struct priv *priv = dev->data->dev_private; + int err; + + priv_lock(priv); + err = priv_set_link(priv, 0); + priv_unlock(priv); + return err; +} + +/** + * DPDK callback to bring the link UP. + * + * @param dev + * Pointer to Ethernet device structure. + * + * @return + * 0 on success, errno value on failure. + */ +static int +mlx4_set_link_up(struct rte_eth_dev *dev) +{ + struct priv *priv = dev->data->dev_private; + int err; + + priv_lock(priv); + err = priv_set_link(priv, 1); + priv_unlock(priv); + return err; +} /** * DPDK callback to get information about the device. * @@ -3818,7 +4140,12 @@ mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) { struct priv *priv = dev->data->dev_private; unsigned int max; + char ifname[IF_NAMESIZE]; + + info->pci_dev = RTE_ETH_DEV_TO_PCI(dev); + if (priv == NULL) + return; priv_lock(priv); /* FIXME: we should ask the device for these values. */ info->min_rx_bufsize = 32; @@ -3834,7 +4161,8 @@ mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) max = 65535; info->max_rx_queues = max; info->max_tx_queues = max; - info->max_mac_addrs = elemof(priv->mac); + /* Last array entry is reserved for broadcast. */ + info->max_mac_addrs = (elemof(priv->mac) - 1); info->rx_offload_capa = (priv->hw_csum ? (DEV_RX_OFFLOAD_IPV4_CKSUM | @@ -3847,12 +4175,38 @@ mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) DEV_TX_OFFLOAD_UDP_CKSUM | DEV_TX_OFFLOAD_TCP_CKSUM) : 0); + if (priv_get_ifname(priv, &ifname) == 0) + info->if_index = if_nametoindex(ifname); + info->speed_capa = + ETH_LINK_SPEED_1G | + ETH_LINK_SPEED_10G | + ETH_LINK_SPEED_20G | + ETH_LINK_SPEED_40G | + ETH_LINK_SPEED_56G; priv_unlock(priv); } -/** - * DPDK callback to get device statistics. - * +static const uint32_t * +mlx4_dev_supported_ptypes_get(struct rte_eth_dev *dev) +{ + static const uint32_t ptypes[] = { + /* refers to rxq_cq_to_pkt_type() */ + RTE_PTYPE_L3_IPV4, + RTE_PTYPE_L3_IPV6, + RTE_PTYPE_INNER_L3_IPV4, + RTE_PTYPE_INNER_L3_IPV6, + RTE_PTYPE_UNKNOWN + }; + + if (dev->rx_pkt_burst == mlx4_rx_burst || + dev->rx_pkt_burst == mlx4_rx_burst_sp) + return ptypes; + return NULL; +} + +/** + * DPDK callback to get device statistics. + * * @param dev * Pointer to Ethernet device structure. * @param[out] stats @@ -3866,6 +4220,8 @@ mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) unsigned int i; unsigned int idx; + if (priv == NULL) + return; priv_lock(priv); /* Add software counters. */ for (i = 0; (i != priv->rxqs_n); ++i) { @@ -3875,17 +4231,13 @@ mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) continue; idx = rxq->stats.idx; if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) { -#ifdef MLX4_PMD_SOFT_COUNTERS tmp.q_ipackets[idx] += rxq->stats.ipackets; tmp.q_ibytes[idx] += rxq->stats.ibytes; -#endif tmp.q_errors[idx] += (rxq->stats.idropped + rxq->stats.rx_nombuf); } -#ifdef MLX4_PMD_SOFT_COUNTERS tmp.ipackets += rxq->stats.ipackets; tmp.ibytes += rxq->stats.ibytes; -#endif tmp.ierrors += rxq->stats.idropped; tmp.rx_nombuf += rxq->stats.rx_nombuf; } @@ -3896,21 +4248,14 @@ mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) continue; idx = txq->stats.idx; if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) { -#ifdef MLX4_PMD_SOFT_COUNTERS tmp.q_opackets[idx] += txq->stats.opackets; tmp.q_obytes[idx] += txq->stats.obytes; -#endif tmp.q_errors[idx] += txq->stats.odropped; } -#ifdef MLX4_PMD_SOFT_COUNTERS tmp.opackets += txq->stats.opackets; tmp.obytes += txq->stats.obytes; -#endif tmp.oerrors += txq->stats.odropped; } -#ifndef MLX4_PMD_SOFT_COUNTERS - /* FIXME: retrieve and add hardware counters. */ -#endif *stats = tmp; priv_unlock(priv); } @@ -3928,6 +4273,8 @@ mlx4_stats_reset(struct rte_eth_dev *dev) unsigned int i; unsigned int idx; + if (priv == NULL) + return; priv_lock(priv); for (i = 0; (i != priv->rxqs_n); ++i) { if ((*priv->rxqs)[i] == NULL) @@ -3939,13 +4286,10 @@ mlx4_stats_reset(struct rte_eth_dev *dev) for (i = 0; (i != priv->txqs_n); ++i) { if ((*priv->txqs)[i] == NULL) continue; - idx = (*priv->rxqs)[i]->stats.idx; + idx = (*priv->txqs)[i]->stats.idx; (*priv->txqs)[i]->stats = (struct mlx4_txq_stats){ .idx = idx }; } -#ifndef MLX4_PMD_SOFT_COUNTERS - /* FIXME: reset hardware counters. */ -#endif priv_unlock(priv); } @@ -3963,13 +4307,12 @@ mlx4_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index) struct priv *priv = dev->data->dev_private; priv_lock(priv); + if (priv->isolated) + goto end; DEBUG("%p: removing MAC address from index %" PRIu32, (void *)dev, index); - if (index >= MLX4_MAX_MAC_ADDRESSES) - goto end; - /* Refuse to remove the broadcast address, this one is special. */ - if (!memcmp(priv->mac[index].addr_bytes, "\xff\xff\xff\xff\xff\xff", - ETHER_ADDR_LEN)) + /* Last array entry is reserved for broadcast. */ + if (index >= (elemof(priv->mac) - 1)) goto end; priv_mac_addr_del(priv, index); end: @@ -3988,27 +4331,50 @@ 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; (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); - if (index >= MLX4_MAX_MAC_ADDRESSES) - goto end; - /* Refuse to add the broadcast address, this one is special. */ - if (!memcmp(mac_addr->addr_bytes, "\xff\xff\xff\xff\xff\xff", - ETHER_ADDR_LEN)) + /* Last array entry is reserved for broadcast. */ + 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; +} + +/** + * DPDK callback to set the primary MAC address. + * + * @param dev + * Pointer to Ethernet device structure. + * @param mac_addr + * MAC address to register. + */ +static void +mlx4_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr) +{ + DEBUG("%p: setting primary MAC address", (void *)dev); + mlx4_mac_addr_remove(dev, 0); + mlx4_mac_addr_add(dev, mac_addr, 0, 0); } /** @@ -4025,6 +4391,12 @@ mlx4_promiscuous_enable(struct rte_eth_dev *dev) int ret; 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; @@ -4033,7 +4405,7 @@ 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; @@ -4071,12 +4443,12 @@ mlx4_promiscuous_disable(struct rte_eth_dev *dev) unsigned int i; 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) @@ -4101,6 +4473,12 @@ mlx4_allmulticast_enable(struct rte_eth_dev *dev) int ret; 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; @@ -4109,7 +4487,7 @@ 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; @@ -4147,12 +4525,12 @@ mlx4_allmulticast_disable(struct rte_eth_dev *dev) unsigned int i; 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) @@ -4164,7 +4542,7 @@ end: } /** - * DPDK callback to retrieve physical link information (unlocked version). + * DPDK callback to retrieve physical link information. * * @param dev * Pointer to Ethernet device structure. @@ -4172,9 +4550,9 @@ end: * Wait for request completion (ignored). */ static int -mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) +mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete) { - struct priv *priv = dev->data->dev_private; + const struct priv *priv = dev->data->dev_private; struct ethtool_cmd edata = { .cmd = ETHTOOL_GSET }; @@ -4182,6 +4560,10 @@ mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) struct rte_eth_link dev_link; int link_speed = 0; + /* priv_lock() is not taken to allow concurrent calls. */ + + if (priv == NULL) + return -EINVAL; (void)wait_to_complete; if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) { WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno)); @@ -4190,7 +4572,7 @@ mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) memset(&dev_link, 0, sizeof(dev_link)); dev_link.link_status = ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING)); - ifr.ifr_data = &edata; + ifr.ifr_data = (void *)&edata; if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) { WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s", strerror(errno)); @@ -4203,6 +4585,8 @@ mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) dev_link.link_speed = link_speed; dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ? ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); + dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds & + ETH_LINK_SPEED_FIXED); if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) { /* Link status changed. */ dev->data->dev_link = dev_link; @@ -4212,26 +4596,6 @@ mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) return -1; } -/** - * DPDK callback to retrieve physical link information. - * - * @param dev - * Pointer to Ethernet device structure. - * @param wait_to_complete - * Wait for request completion (ignored). - */ -static int -mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete) -{ - struct priv *priv = dev->data->dev_private; - int ret; - - priv_lock(priv); - ret = mlx4_link_update_unlocked(dev, wait_to_complete); - priv_unlock(priv); - return ret; -} - /** * DPDK callback to change the MTU. * @@ -4279,17 +4643,15 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) for (i = 0; (i != priv->rxqs_n); ++i) { struct rxq *rxq = (*priv->rxqs)[i]; 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)); - sp = (max_frame_len > (rxq->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) { @@ -4300,7 +4662,7 @@ 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); @@ -4341,7 +4703,7 @@ mlx4_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) }; int ret; - ifr.ifr_data = ðpause; + ifr.ifr_data = (void *)ðpause; priv_lock(priv); if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) { ret = errno; @@ -4389,7 +4751,7 @@ mlx4_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) }; int ret; - ifr.ifr_data = ðpause; + ifr.ifr_data = (void *)ðpause; ethpause.autoneg = fc_conf->autoneg; if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) || (fc_conf->mode & RTE_FC_RX_PAUSE)) @@ -4467,7 +4829,7 @@ 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) @@ -4475,7 +4837,7 @@ 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) @@ -4489,7 +4851,7 @@ 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) @@ -4497,7 +4859,7 @@ 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) @@ -4529,16 +4891,70 @@ mlx4_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) int ret; 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; + 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, .dev_stop = mlx4_dev_stop, + .dev_set_link_down = mlx4_set_link_down, + .dev_set_link_up = mlx4_set_link_up, .dev_close = mlx4_dev_close, .promiscuous_enable = mlx4_promiscuous_enable, .promiscuous_disable = mlx4_promiscuous_disable, @@ -4547,33 +4963,22 @@ static const struct eth_dev_ops mlx4_dev_ops = { .link_update = mlx4_link_update, .stats_get = mlx4_stats_get, .stats_reset = mlx4_stats_reset, - .queue_stats_mapping_set = NULL, .dev_infos_get = mlx4_dev_infos_get, + .dev_supported_ptypes_get = mlx4_dev_supported_ptypes_get, .vlan_filter_set = mlx4_vlan_filter_set, - .vlan_tpid_set = NULL, - .vlan_strip_queue_set = NULL, - .vlan_offload_set = NULL, .rx_queue_setup = mlx4_rx_queue_setup, .tx_queue_setup = mlx4_tx_queue_setup, .rx_queue_release = mlx4_rx_queue_release, .tx_queue_release = mlx4_tx_queue_release, - .dev_led_on = NULL, - .dev_led_off = NULL, .flow_ctrl_get = mlx4_dev_get_flow_ctrl, .flow_ctrl_set = mlx4_dev_set_flow_ctrl, - .priority_flow_ctrl_set = NULL, .mac_addr_remove = mlx4_mac_addr_remove, .mac_addr_add = mlx4_mac_addr_add, + .mac_addr_set = mlx4_mac_addr_set, .mtu_set = mlx4_dev_set_mtu, - .udp_tunnel_add = NULL, - .udp_tunnel_del = NULL, - .fdir_add_signature_filter = NULL, - .fdir_update_signature_filter = NULL, - .fdir_remove_signature_filter = NULL, - .fdir_add_perfect_filter = NULL, - .fdir_update_perfect_filter = NULL, - .fdir_remove_perfect_filter = NULL, - .fdir_set_masks = NULL + .filter_ctrl = mlx4_dev_filter_ctrl, + .rx_queue_intr_enable = mlx4_rx_intr_enable, + .rx_queue_intr_disable = mlx4_rx_intr_disable, }; /** @@ -4613,7 +5018,7 @@ 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, @@ -4648,60 +5053,571 @@ priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN]) return 0; } -/* Support up to 32 adapters. */ -static struct { - struct rte_pci_addr pci_addr; /* associated PCI address */ - uint32_t ports; /* physical ports bitfield. */ -} mlx4_dev[32]; +/** + * Retrieve integer value from environment variable. + * + * @param[in] name + * Environment variable name. + * + * @return + * Integer value, 0 if the variable is not set. + */ +static int +mlx4_getenv_int(const char *name) +{ + const char *val = getenv(name); + + if (val == NULL) + return 0; + return atoi(val); +} + +static void +mlx4_dev_link_status_handler(void *); +static void +mlx4_dev_interrupt_handler(void *); /** - * Get device index in mlx4_dev[] from PCI bus address. + * Link/device status handler. * - * @param[in] pci_addr - * PCI bus address to look for. + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * @param events + * Pointer to event flags holder. * * @return - * mlx4_dev[] index on success, -1 on failure. + * Number of events */ static int -mlx4_dev_idx(struct rte_pci_addr *pci_addr) +priv_dev_status_handler(struct priv *priv, struct rte_eth_dev *dev, + uint32_t *events) { - unsigned int i; - int ret = -1; + struct ibv_async_event event; + int port_change = 0; + struct rte_eth_link *link = &dev->data->dev_link; + int ret = 0; - assert(pci_addr != NULL); - for (i = 0; (i != elemof(mlx4_dev)); ++i) { - if ((mlx4_dev[i].pci_addr.domain == pci_addr->domain) && - (mlx4_dev[i].pci_addr.bus == pci_addr->bus) && - (mlx4_dev[i].pci_addr.devid == pci_addr->devid) && - (mlx4_dev[i].pci_addr.function == pci_addr->function)) - return i; - if ((mlx4_dev[i].ports == 0) && (ret == -1)) - ret = i; + *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) && + (priv->intr_conf.lsc == 1)) { + port_change = 1; + 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) + 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 { + *events |= (1 << RTE_ETH_EVENT_INTR_LSC); } return ret; } /** - * Retrieve integer value from environment variable. + * Handle delayed link status event. * - * @param[in] name - * Environment variable name. + * @param arg + * Registered argument. + */ +static void +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); + priv->pending_alarm = 0; + ret = priv_dev_status_handler(priv, dev, &events); + priv_unlock(priv); + if (ret > 0 && events & (1 << RTE_ETH_EVENT_INTR_LSC)) + _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL, + NULL); +} + +/** + * Handle interrupts from the NIC. + * + * @param[in] intr_handle + * Interrupt handler. + * @param cb_arg + * Callback argument. + */ +static void +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; + + priv_lock(priv); + ret = priv_dev_status_handler(priv, dev, &ev); + priv_unlock(priv); + 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")); + } +} + +/** + * Uninstall interrupt handler. * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. * @return - * Integer value, 0 if the variable is not set. + * 0 on success, negative errno value on failure. */ static int -mlx4_getenv_int(const char *name) +priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev) { - const char *val = getenv(name); + int ret; - if (val == NULL) + if (priv->intr_conf.lsc || + priv->intr_conf.rmv) return 0; - return atoi(val); + 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; +} + +/** + * Install interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * @return + * 0 on success, negative errno value on failure. + */ +static int +priv_dev_interrupt_handler_install(struct priv *priv, + struct rte_eth_dev *dev) +{ + int flags; + int rc; + + /* 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; + 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; } -static struct eth_driver mlx4_driver; +/** + * 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; +} + +/** + * Verify and store value for device argument. + * + * @param[in] key + * Key argument to verify. + * @param[in] val + * Value associated with key. + * @param[in, out] conf + * Shared configuration data. + * + * @return + * 0 on success, negative errno value on failure. + */ +static int +mlx4_arg_parse(const char *key, const char *val, struct mlx4_conf *conf) +{ + 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) { + uint32_t ports = rte_log2_u32(conf->ports.present); + + if (tmp >= ports) { + ERROR("port index %lu outside range [0,%" PRIu32 ")", + tmp, ports); + return -EINVAL; + } + if (!(conf->ports.present & (1 << tmp))) { + ERROR("invalid port index %lu", tmp); + return -EINVAL; + } + conf->ports.enabled |= 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, + (int (*)(const char *, + const char *, + void *)) + 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. @@ -4718,36 +5634,27 @@ static struct eth_driver mlx4_driver; * 0 on success, negative errno value on failure. */ static int -mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) +mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) { struct ibv_device **list; struct ibv_device *ibv_dev; int err = 0; struct ibv_context *attr_ctx = NULL; struct ibv_device_attr device_attr; + struct mlx4_conf conf = { + .ports.present = 0, + }; unsigned int vf; - int idx; int i; (void)pci_drv; - assert(pci_drv == &mlx4_driver.pci_drv); - /* Get mlx4_dev[] index. */ - idx = mlx4_dev_idx(&pci_dev->addr); - if (idx == -1) { - ERROR("this driver cannot support any more adapters"); - return -ENOMEM; - } - DEBUG("using driver device index %d", idx); + assert(pci_drv == &mlx4_driver); - /* Save PCI address. */ - mlx4_dev[idx].pci_addr = pci_dev->addr; 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); @@ -4779,11 +5686,11 @@ mlx4_pci_devinit(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; @@ -4791,46 +5698,64 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) ibv_dev = list[i]; DEBUG("device opened"); - if (ibv_query_device(attr_ctx, &device_attr)) + if (ibv_query_device(attr_ctx, &device_attr)) { + err = ENODEV; goto error; + } INFO("%u port(s) detected", device_attr.phys_port_cnt); + conf.ports.present |= (UINT64_C(1) << device_attr.phys_port_cnt) - 1; + if (mlx4_args(pci_dev->device.devargs, &conf)) { + ERROR("failed to process device arguments"); + err = EINVAL; + goto error; + } + /* Use all ports when none are defined */ + if (!conf.ports.enabled) + conf.ports.enabled = conf.ports.present; 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); struct ibv_context *ctx = NULL; struct ibv_port_attr port_attr; struct ibv_pd *pd = NULL; struct priv *priv = NULL; - struct rte_eth_dev *eth_dev; -#ifdef HAVE_EXP_QUERY_DEVICE + struct rte_eth_dev *eth_dev = NULL; struct ibv_exp_device_attr exp_device_attr; -#endif /* HAVE_EXP_QUERY_DEVICE */ struct ether_addr mac; -#ifdef HAVE_EXP_QUERY_DEVICE + /* If port is not enabled, skip. */ + if (!(conf.ports.enabled & (1 << i))) + continue; exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS; -#ifdef RSS_SUPPORT exp_device_attr.comp_mask |= IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ; -#endif /* RSS_SUPPORT */ -#endif /* HAVE_EXP_QUERY_DEVICE */ - DEBUG("using port %u (%08" PRIx32 ")", port, test); + DEBUG("using port %u", port); ctx = ibv_open_device(ibv_dev); - if (ctx == NULL) + if (ctx == NULL) { + err = ENODEV; goto port_error; + } /* Check port status. */ err = ibv_query_port(ctx, port, &port_attr); if (err) { ERROR("port query failed: %s", strerror(err)); + err = ENODEV; + goto port_error; + } + + if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) { + ERROR("port %d is not configured in Ethernet mode", + port); + err = EINVAL; goto port_error; } + if (port_attr.state != IBV_PORT_ACTIVE) - WARN("bad state for port %d: \"%s\" (%d)", - port, ibv_port_state_str(port_attr.state), - port_attr.state); + DEBUG("port %d is not active: \"%s\" (%d)", + port, ibv_port_state_str(port_attr.state), + port_attr.state); /* Allocate protection domain. */ pd = ibv_alloc_pd(ctx); @@ -4840,8 +5765,6 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) goto port_error; } - mlx4_dev[idx].ports |= test; - /* from rte_ethdev.c */ priv = rte_zmalloc("ethdev private structure", sizeof(*priv), @@ -4857,12 +5780,11 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) priv->port = port; priv->pd = pd; priv->mtu = ETHER_MTU; -#ifdef HAVE_EXP_QUERY_DEVICE if (ibv_exp_query_device(ctx, &exp_device_attr)) { ERROR("ibv_exp_query_device() failed"); + err = ENODEV; goto port_error; } -#ifdef RSS_SUPPORT if ((exp_device_attr.exp_device_cap_flags & IBV_EXP_DEVICE_QPG) && (exp_device_attr.exp_device_cap_flags & @@ -4887,7 +5809,6 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) if (priv->hw_rss) DEBUG("maximum RSS indirection table size: %u", exp_device_attr.max_rss_tbl_sz); -#endif /* RSS_SUPPORT */ priv->hw_csum = ((exp_device_attr.exp_device_cap_flags & @@ -4902,7 +5823,6 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) DEBUG("L2 tunnel checksum offloads are %ssupported", (priv->hw_csum_l2tun ? "" : "not ")); -#ifdef INLINE_RECV priv->inl_recv_size = mlx4_getenv_int("MLX4_INLINE_RECV_SIZE"); if (priv->inl_recv_size) { @@ -4926,15 +5846,13 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) INFO("Set inline receive size to %u", priv->inl_recv_size); } -#endif /* INLINE_RECV */ -#endif /* HAVE_EXP_QUERY_DEVICE */ - (void)mlx4_getenv_int; priv->vf = vf; /* Configure the first MAC address by default. */ if (priv_get_mac(priv, &mac.addr_bytes)) { ERROR("cannot get MAC address, is mlx4_en loaded?" " (errno: %s)", strerror(errno)); + err = ENODEV; goto port_error; } INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x", @@ -4946,7 +5864,7 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) claim_zero(priv_mac_addr_add(priv, 0, (const uint8_t (*)[ETHER_ADDR_LEN]) mac.addr_bytes)); - claim_zero(priv_mac_addr_add(priv, 1, + claim_zero(priv_mac_addr_add(priv, (elemof(priv->mac) - 1), &(const uint8_t [ETHER_ADDR_LEN]) { "\xff\xff\xff\xff\xff\xff" })); #ifndef NDEBUG @@ -4970,7 +5888,7 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) snprintf(name, sizeof(name), "%s port %u", ibv_get_device_name(ibv_dev), port); - eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_PCI); + eth_dev = rte_eth_dev_allocate(name); } if (eth_dev == NULL) { ERROR("can not allocate rte ethdev"); @@ -4979,18 +5897,32 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) } eth_dev->data->dev_private = priv; - eth_dev->pci_dev = pci_dev; - eth_dev->driver = &mlx4_driver; - eth_dev->data->rx_mbuf_alloc_failed = 0; - eth_dev->data->mtu = ETHER_MTU; + eth_dev->data->mac_addrs = priv->mac; + eth_dev->device = &pci_dev->device; + + rte_eth_copy_pci_info(eth_dev, pci_dev); + + 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; - eth_dev->data->mac_addrs = priv->mac; + eth_dev->data->dev_flags |= RTE_ETH_DEV_DETACHABLE; /* 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: @@ -4999,8 +5931,12 @@ port_error: claim_zero(ibv_dealloc_pd(pd)); if (ctx) claim_zero(ibv_close_device(ctx)); + if (eth_dev) + rte_eth_dev_release_port(eth_dev); break; } + if (i == device_attr.phys_port_cnt) + return 0; /* * XXX if something went wrong in the loop above, there is a resource @@ -5009,12 +5945,6 @@ port_error: * way to enumerate the registered ethdevs to free the previous ones. */ - /* no port found, complain */ - if (!mlx4_dev[idx].ports) { - err = ENODEV; - goto error; - } - error: if (attr_ctx) claim_zero(ibv_close_device(attr_ctx)); @@ -5026,45 +5956,40 @@ error: static const struct rte_pci_id mlx4_pci_id_map[] = { { - .vendor_id = PCI_VENDOR_ID_MELLANOX, - .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX3, - .subsystem_vendor_id = PCI_ANY_ID, - .subsystem_device_id = PCI_ANY_ID + RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, + PCI_DEVICE_ID_MELLANOX_CONNECTX3) }, { - .vendor_id = PCI_VENDOR_ID_MELLANOX, - .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX3PRO, - .subsystem_vendor_id = PCI_ANY_ID, - .subsystem_device_id = PCI_ANY_ID + RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, + PCI_DEVICE_ID_MELLANOX_CONNECTX3PRO) }, { - .vendor_id = PCI_VENDOR_ID_MELLANOX, - .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX3VF, - .subsystem_vendor_id = PCI_ANY_ID, - .subsystem_device_id = PCI_ANY_ID + RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, + PCI_DEVICE_ID_MELLANOX_CONNECTX3VF) }, { .vendor_id = 0 } }; -static struct eth_driver mlx4_driver = { - .pci_drv = { - .name = MLX4_DRIVER_NAME, - .id_table = mlx4_pci_id_map, - .devinit = mlx4_pci_devinit, +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, }; /** * Driver initialization routine. */ -static int -rte_mlx4_pmd_init(const char *name, const char *args) +RTE_INIT(rte_mlx4_pmd_init); +static void +rte_mlx4_pmd_init(void) { - (void)name; - (void)args; + RTE_BUILD_BUG_ON(sizeof(wr_id_t) != sizeof(uint64_t)); /* * RDMAV_HUGEPAGES_SAFE tells ibv_fork_init() we intend to use * huge pages. Calling ibv_fork_init() during init allows @@ -5073,14 +5998,10 @@ rte_mlx4_pmd_init(const char *name, const char *args) */ setenv("RDMAV_HUGEPAGES_SAFE", "1", 1); ibv_fork_init(); - rte_eal_pci_register(&mlx4_driver.pci_drv); - return 0; + rte_pci_register(&mlx4_driver); } -static struct rte_driver rte_mlx4_driver = { - .type = PMD_PDEV, - .name = MLX4_DRIVER_NAME, - .init = rte_mlx4_pmd_init, -}; - -PMD_REGISTER_DRIVER(rte_mlx4_driver) +RTE_PMD_EXPORT_NAME(net_mlx4, __COUNTER__); +RTE_PMD_REGISTER_PCI_TABLE(net_mlx4, mlx4_pci_id_map); +RTE_PMD_REGISTER_KMOD_DEP(net_mlx4, + "* ib_uverbs & mlx4_en & mlx4_core & mlx4_ib");