X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx4%2Fmlx4.c;h=6efea24381413f1cc58c477fea65fc18a88b8a76;hb=6751f6deb798394da41467ee6e4cbd10d78ff156;hp=5391b7afdcf7c893ddc86c03eb772439f51fbc67;hpb=82be8d544253a4b5c49b778babf717e5e63f3dc1;p=dpdk.git diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c index 5391b7afdc..6efea24381 100644 --- a/drivers/net/mlx4/mlx4.c +++ b/drivers/net/mlx4/mlx4.c @@ -33,8 +33,6 @@ /* * Known limitations: - * - Multiple RX VLAN filters can be configured, but only the first one - * works properly. * - RSS hash key and options cannot be modified. * - Hardware counters aren't implemented. */ @@ -56,9 +54,9 @@ #include #include #include -#include #include #include +#include /* Verbs header. */ /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */ @@ -74,7 +72,6 @@ #ifdef PEDANTIC #pragma GCC diagnostic ignored "-pedantic" #endif -#include #include #include #include @@ -87,6 +84,8 @@ #include #include #include +#include +#include #ifdef PEDANTIC #pragma GCC diagnostic error "-pedantic" #endif @@ -132,23 +131,11 @@ 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; -} - -/* If raw send operations are available, use them since they are faster. */ -#ifdef SEND_RAW_WR_SUPPORT -typedef struct ibv_send_wr_raw mlx4_send_wr_t; -#define mlx4_post_send ibv_post_send_raw -#else -typedef struct ibv_send_wr mlx4_send_wr_t; -#define mlx4_post_send ibv_post_send -#endif +/* 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. */ @@ -190,31 +177,33 @@ struct rxq { 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. */ /* - * There is exactly one flow configured per MAC address. Each flow - * may contain several specifications, one per configured VLAN ID. + * Each VLAN ID requires a separate flow steering rule. */ BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES); - struct ibv_exp_flow *mac_flow[MLX4_MAX_MAC_ADDRESSES]; - struct ibv_exp_flow *promisc_flow; /* Promiscuous flow. */ - struct ibv_exp_flow *allmulti_flow; /* Multicast flow. */ + 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. */ - uint32_t mb_len; /* Length of a mp-issued mbuf. */ + unsigned int csum:1; /* Enable checksum offloading. */ + unsigned int csum_l2tun:1; /* Same for L2 tunnels. */ struct mlx4_rxq_stats stats; /* RX queue counters. */ unsigned int socket; /* CPU socket ID for allocations. */ + struct ibv_exp_res_domain *rd; /* Resource Domain. */ }; /* TX element. */ struct txq_elt { - mlx4_send_wr_t wr; /* Work Request. */ - struct ibv_sge sges[MLX4_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */ - /* mbuf pointer is derived from WR_ID(wr.wr_id).offset. */ + struct rte_mbuf *buf; }; /* Linear buffer type. It is used when transmitting buffers with too many @@ -229,12 +218,14 @@ typedef uint8_t linear_t[16384]; struct txq { struct priv *priv; /* Back pointer to private data. */ struct { - struct rte_mempool *mp; /* Cached Memory Pool. */ + const struct rte_mempool *mp; /* Cached Memory Pool. */ struct ibv_mr *mr; /* Memory Region (for mp). */ uint32_t lkey; /* mr->lkey */ } mp2mr[MLX4_PMD_TX_MP_CACHE]; /* MP to MR translation table. */ struct ibv_cq *cq; /* Completion Queue. */ struct ibv_qp *qp; /* Queue Pair. */ + struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */ + struct ibv_exp_cq_family *if_cq; /* CQ interface. */ #if MLX4_PMD_MAX_INLINE > 0 uint32_t max_inline; /* Max inline send size <= MLX4_PMD_MAX_INLINE. */ #endif @@ -243,17 +234,19 @@ struct txq { 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. */ }; 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_port_attr port_attr; /* Physical port properties. */ struct ibv_pd *pd; /* Protection Domain. */ /* * MAC addresses array and configuration bit-field. @@ -272,13 +265,15 @@ struct priv { uint8_t port; /* Physical port number. */ unsigned int started:1; /* Device started, flows enabled. */ unsigned int promisc:1; /* Device in promiscuous mode. */ - unsigned int promisc_ok:1; /* Promiscuous flow is supported. */ unsigned int allmulti:1; /* Device receives all multicast packets. */ unsigned int hw_qpg:1; /* QP groups are supported. */ unsigned int hw_tss:1; /* TSS is supported. */ unsigned int hw_rss:1; /* RSS is supported. */ + unsigned int hw_csum:1; /* Checksum offload is supported. */ + unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */ unsigned int rss:1; /* RSS is enabled. */ unsigned int vf:1; /* This is a VF device. */ + unsigned int pending_alarm:1; /* An alarm is pending. */ #ifdef INLINE_RECV unsigned int inl_recv_size; /* Inline recv size */ #endif @@ -289,9 +284,50 @@ struct priv { unsigned int txqs_n; /* TX queues array size. */ struct rxq *(*rxqs)[]; /* RX queues. */ struct txq *(*txqs)[]; /* TX queues. */ + struct rte_intr_handle intr_handle; /* Interrupt handler. */ rte_spinlock_t lock; /* Lock for control functions. */ }; +/* Local storage for secondary process data. */ +struct mlx4_secondary_data { + struct rte_eth_dev_data data; /* Local device data. */ + struct priv *primary_priv; /* Private structure from primary. */ + struct rte_eth_dev_data *shared_dev_data; /* Shared device data. */ + rte_spinlock_t lock; /* Port configuration lock. */ +} mlx4_secondary_data[RTE_MAX_ETHPORTS]; + +/** + * Check if running as a secondary process. + * + * @return + * Nonzero if running as a secondary process. + */ +static inline int +mlx4_is_secondary(void) +{ + return rte_eal_process_type() != RTE_PROC_PRIMARY; +} + +/** + * Return private structure associated with an Ethernet device. + * + * @param dev + * Pointer to Ethernet device structure. + * + * @return + * Pointer to private structure. + */ +static struct priv * +mlx4_get_priv(struct rte_eth_dev *dev) +{ + struct mlx4_secondary_data *sd; + + if (!mlx4_is_secondary()) + return dev->data->dev_private; + sd = &mlx4_secondary_data[dev->data->port_id]; + return sd->data.dev_private; +} + /** * Lock private structure to protect it from concurrent access in the * control path. @@ -337,9 +373,11 @@ priv_unlock(struct priv *priv) static int priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE]) { - int ret = -1; DIR *dir; struct dirent *dent; + unsigned int dev_type = 0; + unsigned int dev_port_prev = ~0u; + char match[IF_NAMESIZE] = ""; { MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path); @@ -351,7 +389,7 @@ priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE]) while ((dent = readdir(dir)) != NULL) { char *name = dent->d_name; FILE *file; - unsigned int dev_id; + unsigned int dev_port; int r; if ((name[0] == '.') && @@ -359,22 +397,47 @@ priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE]) ((name[1] == '.') && (name[2] == '\0')))) continue; - MKSTR(path, "%s/device/net/%s/dev_id", - priv->ctx->device->ibdev_path, name); + MKSTR(path, "%s/device/net/%s/%s", + priv->ctx->device->ibdev_path, name, + (dev_type ? "dev_id" : "dev_port")); file = fopen(path, "rb"); - if (file == NULL) + if (file == NULL) { + if (errno != ENOENT) + continue; + /* + * Switch to dev_id when dev_port does not exist as + * is the case with Linux kernel versions < 3.15. + */ +try_dev_id: + match[0] = '\0'; + if (dev_type) + break; + dev_type = 1; + dev_port_prev = ~0u; + rewinddir(dir); continue; - r = fscanf(file, "%x", &dev_id); - fclose(file); - if ((r == 1) && (dev_id == (priv->port - 1u))) { - snprintf(*ifname, sizeof(*ifname), "%s", name); - ret = 0; - break; } + r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port); + fclose(file); + if (r != 1) + continue; + /* + * Switch to dev_id when dev_port returns the same value for + * all ports. May happen when using a MOFED release older than + * 3.0 with a Linux kernel >= 3.15. + */ + if (dev_port == dev_port_prev) + goto try_dev_id; + dev_port_prev = dev_port; + if (dev_port == (priv->port - 1u)) + snprintf(match, sizeof(match), "%s", name); } closedir(dir); - return ret; + if (match[0] == '\0') + return -1; + strncpy(*ifname, match, sizeof(*ifname)); + return 0; } /** @@ -594,7 +657,15 @@ 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; } /** @@ -618,15 +689,22 @@ 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, + unsigned int socket, int inactive, const struct rte_eth_rxconf *conf, struct rte_mempool *mp); static void @@ -662,6 +740,15 @@ dev_configure(struct rte_eth_dev *dev) } if (rxqs_n == priv->rxqs_n) return 0; + if (!rte_is_power_of_2(rxqs_n)) { + 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. */ @@ -697,7 +784,7 @@ 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); + ret = rxq_setup(dev, &priv->rxq_parent, 0, 0, 0, NULL, NULL); if (!ret) return 0; /* Failure, rollback. */ @@ -722,6 +809,8 @@ mlx4_dev_configure(struct rte_eth_dev *dev) struct priv *priv = dev->data->dev_private; int ret; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; priv_lock(priv); ret = dev_configure(dev); assert(ret >= 0); @@ -729,6 +818,160 @@ 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); + +/** + * Configure secondary process queues from a private data pointer (primary + * or secondary) and update burst callbacks. Can take place only once. + * + * All queues must have been previously created by the primary process to + * avoid undefined behavior. + * + * @param priv + * Private data pointer from either primary or secondary process. + * + * @return + * Private data pointer from secondary process, NULL in case of error. + */ +static struct priv * +mlx4_secondary_data_setup(struct priv *priv) +{ + unsigned int port_id = 0; + struct mlx4_secondary_data *sd; + void **tx_queues; + void **rx_queues; + unsigned int nb_tx_queues; + unsigned int nb_rx_queues; + unsigned int i; + + /* priv must be valid at this point. */ + assert(priv != NULL); + /* priv->dev must also be valid but may point to local memory from + * another process, possibly with the same address and must not + * be dereferenced yet. */ + assert(priv->dev != NULL); + /* Determine port ID by finding out where priv comes from. */ + while (1) { + sd = &mlx4_secondary_data[port_id]; + rte_spinlock_lock(&sd->lock); + /* Primary process? */ + if (sd->primary_priv == priv) + break; + /* Secondary process? */ + if (sd->data.dev_private == priv) + break; + rte_spinlock_unlock(&sd->lock); + if (++port_id == RTE_DIM(mlx4_secondary_data)) + port_id = 0; + } + /* Switch to secondary private structure. If private data has already + * been updated by another thread, there is nothing else to do. */ + priv = sd->data.dev_private; + if (priv->dev->data == &sd->data) + goto end; + /* Sanity checks. Secondary private structure is supposed to point + * to local eth_dev, itself still pointing to the shared device data + * structure allocated by the primary process. */ + assert(sd->shared_dev_data != &sd->data); + assert(sd->data.nb_tx_queues == 0); + assert(sd->data.tx_queues == NULL); + assert(sd->data.nb_rx_queues == 0); + assert(sd->data.rx_queues == NULL); + assert(priv != sd->primary_priv); + assert(priv->dev->data == sd->shared_dev_data); + assert(priv->txqs_n == 0); + assert(priv->txqs == NULL); + assert(priv->rxqs_n == 0); + assert(priv->rxqs == NULL); + nb_tx_queues = sd->shared_dev_data->nb_tx_queues; + nb_rx_queues = sd->shared_dev_data->nb_rx_queues; + /* Allocate local storage for queues. */ + tx_queues = rte_zmalloc("secondary ethdev->tx_queues", + sizeof(sd->data.tx_queues[0]) * nb_tx_queues, + RTE_CACHE_LINE_SIZE); + rx_queues = rte_zmalloc("secondary ethdev->rx_queues", + sizeof(sd->data.rx_queues[0]) * nb_rx_queues, + RTE_CACHE_LINE_SIZE); + if (tx_queues == NULL || rx_queues == NULL) + goto error; + /* Lock to prevent control operations during setup. */ + priv_lock(priv); + /* TX queues. */ + for (i = 0; i != nb_tx_queues; ++i) { + struct txq *primary_txq = (*sd->primary_priv->txqs)[i]; + struct txq *txq; + + if (primary_txq == NULL) + continue; + txq = rte_calloc_socket("TXQ", 1, sizeof(*txq), 0, + primary_txq->socket); + if (txq != NULL) { + if (txq_setup(priv->dev, + txq, + primary_txq->elts_n * MLX4_PMD_SGE_WR_N, + primary_txq->socket, + NULL) == 0) { + txq->stats.idx = primary_txq->stats.idx; + tx_queues[i] = txq; + continue; + } + rte_free(txq); + } + while (i) { + txq = tx_queues[--i]; + txq_cleanup(txq); + rte_free(txq); + } + goto error; + } + /* RX queues. */ + for (i = 0; i != nb_rx_queues; ++i) { + struct rxq *primary_rxq = (*sd->primary_priv->rxqs)[i]; + + if (primary_rxq == NULL) + continue; + /* Not supported yet. */ + rx_queues[i] = NULL; + } + /* Update everything. */ + priv->txqs = (void *)tx_queues; + priv->txqs_n = nb_tx_queues; + priv->rxqs = (void *)rx_queues; + priv->rxqs_n = nb_rx_queues; + sd->data.rx_queues = rx_queues; + sd->data.tx_queues = tx_queues; + sd->data.nb_rx_queues = nb_rx_queues; + sd->data.nb_tx_queues = nb_tx_queues; + sd->data.dev_link = sd->shared_dev_data->dev_link; + sd->data.mtu = sd->shared_dev_data->mtu; + memcpy(sd->data.rx_queue_state, sd->shared_dev_data->rx_queue_state, + sizeof(sd->data.rx_queue_state)); + memcpy(sd->data.tx_queue_state, sd->shared_dev_data->tx_queue_state, + sizeof(sd->data.tx_queue_state)); + sd->data.dev_flags = sd->shared_dev_data->dev_flags; + /* Use local data from now on. */ + rte_mb(); + priv->dev->data = &sd->data; + rte_mb(); + priv->dev->tx_pkt_burst = mlx4_tx_burst; + priv->dev->rx_pkt_burst = removed_rx_burst; + priv_unlock(priv); +end: + /* More sanity checks. */ + assert(priv->dev->tx_pkt_burst == mlx4_tx_burst); + assert(priv->dev->rx_pkt_burst == removed_rx_burst); + assert(priv->dev->data == &sd->data); + rte_spinlock_unlock(&sd->lock); + return priv; +error: + priv_unlock(priv); + rte_free(tx_queues); + rte_free(rx_queues); + rte_spinlock_unlock(&sd->lock); + return NULL; +} + /* TX queues handling. */ /** @@ -761,7 +1004,7 @@ 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); @@ -770,14 +1013,8 @@ txq_alloc_elts(struct txq *txq, unsigned int elts_n) } for (i = 0; (i != elts_n); ++i) { struct txq_elt *elt = &(*elts)[i]; - mlx4_send_wr_t *wr = &elt->wr; - /* Configure WR. */ - WR_ID(wr->wr_id).id = i; - WR_ID(wr->wr_id).offset = 0; - wr->sg_list = &elt->sges[0]; - wr->opcode = IBV_WR_SEND; - /* Other fields are updated during TX. */ + elt->buf = NULL; } DEBUG("%p: allocated and configured %u WRs", (void *)txq, elts_n); txq->elts_n = elts_n; @@ -785,6 +1022,12 @@ txq_alloc_elts(struct txq *txq, unsigned int elts_n) txq->elts_head = 0; txq->elts_tail = 0; txq->elts_comp = 0; + /* Request send completion every MLX4_PMD_TX_PER_COMP_REQ packets or + * at least 4 times per ring. */ + txq->elts_comp_cd_init = + ((MLX4_PMD_TX_PER_COMP_REQ < (elts_n / 4)) ? + MLX4_PMD_TX_PER_COMP_REQ : (elts_n / 4)); + txq->elts_comp_cd = txq->elts_comp_cd_init; txq->elts_linear = elts_linear; txq->mr_linear = mr_linear; assert(ret == 0); @@ -810,14 +1053,20 @@ 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; @@ -827,13 +1076,17 @@ 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 (WR_ID(elt->wr.wr_id).offset == 0) - continue; - rte_pktmbuf_free((void *)((uintptr_t)elt->sges[0].addr - - WR_ID(elt->wr.wr_id).offset)); + 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); } @@ -850,14 +1103,48 @@ txq_free_elts(struct txq *txq) static void txq_cleanup(struct txq *txq) { + struct ibv_exp_release_intf_params params; size_t i; DEBUG("cleaning up %p", (void *)txq); txq_free_elts(txq); + if (txq->if_qp != NULL) { + assert(txq->priv != NULL); + assert(txq->priv->ctx != NULL); + assert(txq->qp != NULL); + params = (struct ibv_exp_release_intf_params){ + .comp_mask = 0, + }; + claim_zero(ibv_exp_release_intf(txq->priv->ctx, + txq->if_qp, + ¶ms)); + } + if (txq->if_cq != NULL) { + assert(txq->priv != NULL); + assert(txq->priv->ctx != NULL); + assert(txq->cq != NULL); + params = (struct ibv_exp_release_intf_params){ + .comp_mask = 0, + }; + claim_zero(ibv_exp_release_intf(txq->priv->ctx, + txq->if_cq, + ¶ms)); + } if (txq->qp != NULL) claim_zero(ibv_destroy_qp(txq->qp)); if (txq->cq != NULL) claim_zero(ibv_destroy_cq(txq->cq)); + if (txq->rd != NULL) { + struct ibv_exp_destroy_res_domain_attr attr = { + .comp_mask = 0, + }; + + assert(txq->priv != NULL); + assert(txq->priv->ctx != NULL); + claim_zero(ibv_exp_destroy_res_domain(txq->priv->ctx, + txq->rd, + &attr)); + } for (i = 0; (i != elemof(txq->mp2mr)); ++i) { if (txq->mp2mr[i].mp == NULL) break; @@ -871,9 +1158,9 @@ txq_cleanup(struct txq *txq) * Manage TX completions. * * When sending a burst, mlx4_tx_burst() posts several WRs. - * To improve performance, a completion event is only required for the last of - * them. Doing so discards completion information for other WRs, but this - * information would not be used anyway. + * To improve performance, a completion event is only required once every + * MLX4_PMD_TX_PER_COMP_REQ sends. Doing so discards completion information + * for other WRs, but this information would not be used anyway. * * @param txq * Pointer to TX queue structure. @@ -885,9 +1172,8 @@ static int txq_complete(struct txq *txq) { unsigned int elts_comp = txq->elts_comp; - unsigned int elts_tail; + unsigned int elts_tail = txq->elts_tail; const unsigned int elts_n = txq->elts_n; - struct ibv_wc wcs[elts_comp]; int wcs_n; if (unlikely(elts_comp == 0)) @@ -896,7 +1182,7 @@ txq_complete(struct txq *txq) DEBUG("%p: processing %u work requests completions", (void *)txq, elts_comp); #endif - wcs_n = ibv_poll_cq(txq->cq, elts_comp, wcs); + wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp); if (unlikely(wcs_n == 0)) return 0; if (unlikely(wcs_n < 0)) { @@ -907,22 +1193,150 @@ txq_complete(struct txq *txq) elts_comp -= wcs_n; assert(elts_comp <= txq->elts_comp); /* - * Work Completion ID contains the associated element index in - * (*txq->elts)[]. Since WCs are returned in order, we only need to - * look at the last WC to clear older Work Requests. - * * Assume WC status is successful as nothing can be done about it * anyway. */ - elts_tail = WR_ID(wcs[wcs_n - 1].wr_id).id; - /* Consume the last WC. */ - if (++elts_tail >= elts_n) - elts_tail = 0; + elts_tail += wcs_n * txq->elts_comp_cd_init; + if (elts_tail >= elts_n) + elts_tail -= elts_n; txq->elts_tail = elts_tail; txq->elts_comp = elts_comp; 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 *) + __attribute__((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, @@ -954,11 +1368,9 @@ 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); @@ -969,7 +1381,7 @@ 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]))); } @@ -977,11 +1389,68 @@ 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; } +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); +} + +#if MLX4_PMD_SGE_WR_N > 1 + /** * Copy scattered mbuf contents to a single linear buffer. * @@ -1014,6 +1483,122 @@ linearize_mbuf(linear_t *linear, struct rte_mbuf *buf) return size; } +/** + * Handle scattered buffers for mlx4_tx_burst(). + * + * @param txq + * TX queue structure. + * @param segs + * Number of segments in buf. + * @param elt + * TX queue element to fill. + * @param[in] buf + * Buffer to process. + * @param elts_head + * Index of the linear buffer to use if necessary (normally txq->elts_head). + * @param[out] sges + * Array filled with SGEs on success. + * + * @return + * A structure containing the processed packet size in bytes and the + * number of SGEs. Both fields are set to (unsigned int)-1 in case of + * failure. + */ +static struct tx_burst_sg_ret { + unsigned int length; + unsigned int num; +} +tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt, + struct rte_mbuf *buf, unsigned int elts_head, + struct ibv_sge (*sges)[MLX4_PMD_SGE_WR_N]) +{ + unsigned int sent_size = 0; + unsigned int j; + int linearize = 0; + + /* When there are too many segments, extra segments are + * linearized in the last SGE. */ + if (unlikely(segs > elemof(*sges))) { + segs = (elemof(*sges) - 1); + linearize = 1; + } + /* Update element. */ + elt->buf = buf; + /* Register segments as SGEs. */ + for (j = 0; (j != segs); ++j) { + struct ibv_sge *sge = &(*sges)[j]; + uint32_t lkey; + + /* Retrieve Memory Region key for this memory 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", + (void *)txq); + /* Clean up TX element. */ + elt->buf = NULL; + goto stop; + } + /* Update SGE. */ + sge->addr = rte_pktmbuf_mtod(buf, uintptr_t); + if (txq->priv->vf) + rte_prefetch0((volatile void *) + (uintptr_t)sge->addr); + sge->length = DATA_LEN(buf); + sge->lkey = lkey; + sent_size += sge->length; + buf = NEXT(buf); + } + /* If buf is not NULL here and is not going to be linearized, + * nb_segs is not valid. */ + assert(j == segs); + assert((buf == NULL) || (linearize)); + /* Linearize extra segments. */ + if (linearize) { + struct ibv_sge *sge = &(*sges)[segs]; + linear_t *linear = &(*txq->elts_linear)[elts_head]; + unsigned int size = linearize_mbuf(linear, buf); + + assert(segs == (elemof(*sges) - 1)); + if (size == 0) { + /* Invalid packet. */ + DEBUG("%p: packet too large to be linearized.", + (void *)txq); + /* Clean up TX element. */ + elt->buf = NULL; + goto stop; + } + /* If MLX4_PMD_SGE_WR_N is 1, free mbuf immediately. */ + if (elemof(*sges) == 1) { + do { + struct rte_mbuf *next = NEXT(buf); + + rte_pktmbuf_free_seg(buf); + buf = next; + } while (buf != NULL); + elt->buf = NULL; + } + /* Update SGE. */ + sge->addr = (uintptr_t)&(*linear)[0]; + 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, + .num = segs, + }; +stop: + return (struct tx_burst_sg_ret){ + .length = -1, + .num = -1, + }; +} + +#endif /* MLX4_PMD_SGE_WR_N > 1 */ + /** * DPDK callback for TX. * @@ -1031,18 +1616,17 @@ static uint16_t mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) { struct txq *txq = (struct txq *)dpdk_txq; - mlx4_send_wr_t head; - mlx4_send_wr_t **wr_next = &head.next; - mlx4_send_wr_t *bad_wr; 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; unsigned int i; unsigned int max; int err; + 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); @@ -1055,21 +1639,24 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) max = pkts_n; for (i = 0; (i != max); ++i) { struct rte_mbuf *buf = pkts[i]; + unsigned int elts_head_next = + (((elts_head + 1) == elts_n) ? 0 : elts_head + 1); + struct txq_elt *elt_next = &(*txq->elts)[elts_head_next]; struct txq_elt *elt = &(*txq->elts)[elts_head]; - mlx4_send_wr_t *wr = &elt->wr; unsigned int segs = NB_SEGS(buf); -#if (MLX4_PMD_MAX_INLINE > 0) || defined(MLX4_PMD_SOFT_COUNTERS) +#ifdef MLX4_PMD_SOFT_COUNTERS unsigned int sent_size = 0; #endif - unsigned int j; - int linearize = 0; + uint32_t send_flags = 0; /* Clean up old buffer. */ - if (likely(WR_ID(wr->wr_id).offset != 0)) { - struct rte_mbuf *tmp = (void *) - ((uintptr_t)elt->sges[0].addr - - WR_ID(wr->wr_id).offset); + 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); @@ -1078,148 +1665,96 @@ mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) tmp = next; } while (tmp != NULL); } -#ifndef NDEBUG - /* For assert(). */ - WR_ID(wr->wr_id).offset = 0; - for (j = 0; ((int)j < wr->num_sge); ++j) { - elt->sges[j].addr = 0; - elt->sges[j].length = 0; - elt->sges[j].lkey = 0; + /* Request TX completion. */ + if (unlikely(--elts_comp_cd == 0)) { + elts_comp_cd = txq->elts_comp_cd_init; + ++elts_comp; + send_flags |= IBV_EXP_QP_BURST_SIGNALED; } - wr->next = NULL; - wr->num_sge = 0; -#endif - /* Sanity checks, most of which are only relevant with - * debugging enabled. */ - assert(WR_ID(wr->wr_id).id == elts_head); - assert(WR_ID(wr->wr_id).offset == 0); - assert(wr->next == NULL); - assert(wr->sg_list == &elt->sges[0]); - assert(wr->num_sge == 0); - assert(wr->opcode == IBV_WR_SEND); - /* When there are too many segments, extra segments are - * linearized in the last SGE. */ - if (unlikely(segs > elemof(elt->sges))) { - segs = (elemof(elt->sges) - 1); - linearize = 1; + /* Should we enable HW CKSUM offload */ + if (buf->ol_flags & + (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) { + send_flags |= IBV_EXP_QP_BURST_IP_CSUM; + /* HW does not support checksum offloads at arbitrary + * offsets but automatically recognizes the packet + * type. For inner L3/L4 checksums, only VXLAN (UDP) + * tunnels are currently supported. */ + if (RTE_ETH_IS_TUNNEL_PKT(buf->packet_type)) + send_flags |= IBV_EXP_QP_BURST_TUNNEL; } - /* Set WR fields. */ - assert((rte_pktmbuf_mtod(buf, uintptr_t) - - (uintptr_t)buf) <= 0xffff); - WR_ID(wr->wr_id).offset = - (rte_pktmbuf_mtod(buf, uintptr_t) - - (uintptr_t)buf); - wr->num_sge = segs; - /* Register segments as SGEs. */ - for (j = 0; (j != segs); ++j) { - struct ibv_sge *sge = &elt->sges[j]; + if (likely(segs == 1)) { + uintptr_t addr; + uint32_t length; uint32_t lkey; + /* Retrieve buffer information. */ + 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" " association", (void *)txq); /* Clean up TX element. */ - WR_ID(elt->wr.wr_id).offset = 0; -#ifndef NDEBUG - /* For assert(). */ - while (j) { - --j; - --sge; - sge->addr = 0; - sge->length = 0; - sge->lkey = 0; - } - wr->num_sge = 0; -#endif + elt->buf = NULL; goto stop; } - /* Sanity checks, only relevant with debugging - * enabled. */ - assert(sge->addr == 0); - assert(sge->length == 0); - assert(sge->lkey == 0); - /* Update SGE. */ - sge->addr = rte_pktmbuf_mtod(buf, uintptr_t); + /* Update element. */ + elt->buf = buf; if (txq->priv->vf) rte_prefetch0((volatile void *) - (uintptr_t)sge->addr); - sge->length = DATA_LEN(buf); - sge->lkey = lkey; -#if (MLX4_PMD_MAX_INLINE > 0) || defined(MLX4_PMD_SOFT_COUNTERS) - sent_size += sge->length; -#endif - buf = NEXT(buf); - } - /* If buf is not NULL here and is not going to be linearized, - * nb_segs is not valid. */ - assert(j == segs); - assert((buf == NULL) || (linearize)); - /* Linearize extra segments. */ - if (linearize) { - struct ibv_sge *sge = &elt->sges[segs]; - linear_t *linear = &(*txq->elts_linear)[elts_head]; - unsigned int size = linearize_mbuf(linear, buf); - - assert(segs == (elemof(elt->sges) - 1)); - if (size == 0) { - /* Invalid packet. */ - DEBUG("%p: packet too large to be linearized.", - (void *)txq); - /* Clean up TX element. */ - WR_ID(elt->wr.wr_id).offset = 0; -#ifndef NDEBUG - /* For assert(). */ - while (j) { - --j; - --sge; - sge->addr = 0; - sge->length = 0; - sge->lkey = 0; - } - wr->num_sge = 0; + (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, + (void *)addr, + length, + send_flags); + else #endif + err = txq->if_qp->send_pending + (txq->qp, + addr, + length, + lkey, + send_flags); + if (unlikely(err)) goto stop; - } - /* If MLX4_PMD_SGE_WR_N is 1, free mbuf immediately - * and clear offset from WR ID. */ - if (elemof(elt->sges) == 1) { - do { - struct rte_mbuf *next = NEXT(buf); +#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; - rte_pktmbuf_free_seg(buf); - buf = next; - } while (buf != NULL); - WR_ID(wr->wr_id).offset = 0; - } - /* Set WR fields and fill SGE with linear buffer. */ - ++wr->num_sge; - /* Sanity checks, only relevant with debugging - * enabled. */ - assert(sge->addr == 0); - assert(sge->length == 0); - assert(sge->lkey == 0); - /* Update SGE. */ - sge->addr = (uintptr_t)&(*linear)[0]; - sge->length = size; - sge->lkey = txq->mr_linear->lkey; -#if (MLX4_PMD_MAX_INLINE > 0) || defined(MLX4_PMD_SOFT_COUNTERS) - sent_size += size; + ret = tx_burst_sg(txq, segs, elt, buf, elts_head, + &sges); + if (ret.length == (unsigned int)-1) + goto stop; + RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf); + /* Put SG list into send queue. */ + err = txq->if_qp->send_pending_sg_list + (txq->qp, + sges, + ret.num, + 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 */ } - /* Link WRs together for ibv_post_send(). */ - *wr_next = wr; - wr_next = &wr->next; -#if MLX4_PMD_MAX_INLINE > 0 - if (sent_size <= txq->max_inline) - wr->send_flags = IBV_SEND_INLINE; - else -#endif - wr->send_flags = 0; - if (++elts_head >= elts_n) - elts_head = 0; + elts_head = elts_head_next; #ifdef MLX4_PMD_SOFT_COUNTERS /* Increment sent bytes counter. */ txq->stats.obytes += sent_size; @@ -1233,64 +1768,60 @@ stop: /* Increment sent packets counter. */ txq->stats.opackets += i; #endif - *wr_next = NULL; - /* The last WR is the only one asking for a completion event. */ - containerof(wr_next, mlx4_send_wr_t, next)-> - send_flags |= IBV_SEND_SIGNALED; - err = mlx4_post_send(txq->qp, head.next, &bad_wr); + /* Ring QP doorbell. */ + err = txq->if_qp->send_flush(txq->qp); if (unlikely(err)) { - unsigned int unsent = 0; - - /* An error occurred, completion event is lost. Fix counters. */ - while (bad_wr != NULL) { - struct txq_elt *elt = - containerof(bad_wr, struct txq_elt, wr); - mlx4_send_wr_t *wr = &elt->wr; - mlx4_send_wr_t *next = wr->next; -#if defined(MLX4_PMD_SOFT_COUNTERS) || !defined(NDEBUG) - unsigned int j; -#endif - - assert(wr == bad_wr); - /* Clean up TX element without freeing it, caller - * should take care of this. */ - WR_ID(elt->wr.wr_id).offset = 0; -#ifdef MLX4_PMD_SOFT_COUNTERS - for (j = 0; ((int)j < wr->num_sge); ++j) - txq->stats.obytes -= wr->sg_list[j].length; -#endif - ++unsent; -#ifndef NDEBUG - /* For assert(). */ - for (j = 0; ((int)j < wr->num_sge); ++j) { - elt->sges[j].addr = 0; - elt->sges[j].length = 0; - elt->sges[j].lkey = 0; - } - wr->next = NULL; - wr->num_sge = 0; -#endif - bad_wr = next; - } -#ifdef MLX4_PMD_SOFT_COUNTERS - txq->stats.opackets -= unsent; -#endif - assert(i >= unsent); - i -= unsent; - /* "Unsend" remaining packets. */ - elts_head -= unsent; - if (elts_head >= elts_n) - elts_head += elts_n; - assert(elts_head < elts_n); - DEBUG("%p: mlx4_post_send() failed, %u unprocessed WRs: %s", - (void *)txq, unsent, - ((err <= -1) ? "Internal error" : strerror(err))); - } else - ++txq->elts_comp; + /* A nonzero value is not supposed to be returned. + * Nothing can be done about it. */ + DEBUG("%p: send_flush() failed with error %d", + (void *)txq, err); + } txq->elts_head = elts_head; + txq->elts_comp += elts_comp; + txq->elts_comp_cd = elts_comp_cd; return i; } +/** + * DPDK callback for TX in secondary processes. + * + * This function configures all queues from primary process information + * if necessary before reverting to the normal TX burst callback. + * + * @param dpdk_txq + * Generic pointer to TX queue structure. + * @param[in] pkts + * Packets to transmit. + * @param pkts_n + * Number of packets in array. + * + * @return + * Number of packets successfully transmitted (<= pkts_n). + */ +static uint16_t +mlx4_tx_burst_secondary_setup(void *dpdk_txq, struct rte_mbuf **pkts, + uint16_t pkts_n) +{ + struct txq *txq = dpdk_txq; + struct priv *priv = mlx4_secondary_data_setup(txq->priv); + struct priv *primary_priv; + unsigned int index; + + if (priv == NULL) + return 0; + primary_priv = + mlx4_secondary_data[priv->dev->data->port_id].primary_priv; + /* Look for queue index in both private structures. */ + for (index = 0; index != priv->txqs_n; ++index) + if (((*primary_priv->txqs)[index] == txq) || + ((*priv->txqs)[index] == txq)) + break; + if (index == priv->txqs_n) + return 0; + txq = (*priv->txqs)[index]; + return priv->dev->tx_pkt_burst(txq, pkts, pkts_n); +} + /** * Configure a TX queue. * @@ -1312,26 +1843,49 @@ static int txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, unsigned int socket, const struct rte_eth_txconf *conf) { - struct priv *priv = dev->data->dev_private; + struct priv *priv = mlx4_get_priv(dev); struct txq tmpl = { .priv = priv, .socket = socket }; union { - struct ibv_qp_init_attr init; + struct ibv_exp_query_intf_params params; + struct ibv_exp_qp_init_attr init; + struct ibv_exp_res_domain_init_attr rd; + struct ibv_exp_cq_init_attr cq; struct ibv_exp_qp_attr mod; } attr; + enum ibv_exp_query_intf_status status; 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, desc); + " multiple of %d)", (void *)dev, MLX4_PMD_SGE_WR_N); return EINVAL; } desc /= MLX4_PMD_SGE_WR_N; /* MRs will be registered in mp2mr[] later. */ - tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, NULL, 0); + attr.rd = (struct ibv_exp_res_domain_init_attr){ + .comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL | + IBV_EXP_RES_DOMAIN_MSG_MODEL), + .thread_model = IBV_EXP_THREAD_SINGLE, + .msg_model = IBV_EXP_MSG_HIGH_BW, + }; + tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd); + if (tmpl.rd == NULL) { + ret = ENOMEM; + ERROR("%p: RD 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); if (tmpl.cq == NULL) { ret = ENOMEM; ERROR("%p: CQ creation failure: %s", @@ -1342,7 +1896,7 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, priv->device_attr.max_qp_wr); DEBUG("priv->device_attr.max_sge is %d", priv->device_attr.max_sge); - attr.init = (struct ibv_qp_init_attr){ + attr.init = (struct ibv_exp_qp_init_attr){ /* CQ to be associated with the send queue. */ .send_cq = tmpl.cq, /* CQ to be associated with the receive queue. */ @@ -1364,9 +1918,13 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, .qp_type = IBV_QPT_RAW_PACKET, /* Do *NOT* enable this, completions events are managed per * TX burst. */ - .sq_sig_all = 0 + .sq_sig_all = 0, + .pd = priv->pd, + .res_domain = tmpl.rd, + .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD | + IBV_EXP_QP_INIT_ATTR_RES_DOMAIN), }; - tmpl.qp = ibv_create_qp(priv->pd, &attr.init); + tmpl.qp = ibv_exp_create_qp(priv->ctx, &attr.init); if (tmpl.qp == NULL) { ret = (errno ? errno : EINVAL); ERROR("%p: QP creation failure: %s", @@ -1412,11 +1970,42 @@ txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc, (void *)dev, strerror(ret)); goto error; } + attr.params = (struct ibv_exp_query_intf_params){ + .intf_scope = IBV_EXP_INTF_GLOBAL, + .intf = IBV_EXP_INTF_CQ, + .obj = tmpl.cq, + }; + tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status); + if (tmpl.if_cq == NULL) { + 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, +#ifdef HAVE_EXP_QP_BURST_CREATE_DISABLE_ETH_LOOPBACK + /* MC loopback must be disabled when not using a VF. */ + .family_flags = + (!priv->vf ? + IBV_EXP_QP_BURST_CREATE_DISABLE_ETH_LOOPBACK : + 0), +#endif + }; + 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 txq in case we're reinitializing it. */ DEBUG("%p: cleaning-up old txq just in case", (void *)txq); 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: @@ -1450,6 +2039,8 @@ mlx4_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, struct txq *txq = (*priv->txqs)[idx]; int ret; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; priv_lock(priv); DEBUG("%p: configuring queue %u for %u descriptors", (void *)dev, idx, desc); @@ -1505,6 +2096,8 @@ mlx4_tx_queue_release(void *dpdk_txq) struct priv *priv; unsigned int i; + if (mlx4_is_secondary()) + return; if (txq == NULL) return; priv = txq->priv; @@ -1613,6 +2206,7 @@ rxq_alloc_elts_sp(struct rxq *rxq, unsigned int elts_n, DEBUG("%p: allocated and configured %u WRs (%zu segments)", (void *)rxq, elts_n, (elts_n * elemof((*elts)[0].sges))); rxq->elts_n = elts_n; + rxq->elts_head = 0; rxq->elts.sp = elts; assert(ret == 0); return 0; @@ -1758,6 +2352,7 @@ rxq_alloc_elts(struct rxq *rxq, unsigned int elts_n, struct rte_mbuf **pool) DEBUG("%p: allocated and configured %u single-segment WRs", (void *)rxq, elts_n); rxq->elts_n = elts_n; + rxq->elts_head = 0; rxq->elts.no_sp = elts; assert(ret == 0); return 0; @@ -1815,15 +2410,17 @@ rxq_free_elts(struct rxq *rxq) } /** - * Unregister a MAC address from a RX queue. + * Delete flow steering rule. * * @param rxq * Pointer to RX queue structure. * @param mac_index * MAC address index. + * @param vlan_index + * VLAN index. */ static void -rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index) +rxq_del_flow(struct rxq *rxq, unsigned int mac_index, unsigned int vlan_index) { #ifndef NDEBUG struct priv *priv = rxq->priv; @@ -1831,20 +2428,43 @@ rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index) (const uint8_t (*)[ETHER_ADDR_LEN]) priv->mac[mac_index].addr_bytes; #endif + assert(rxq->mac_flow[mac_index][vlan_index] != NULL); + DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u" + " (VLAN ID %" PRIu16 ")", + (void *)rxq, + (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5], + mac_index, priv->vlan_filter[vlan_index].id); + claim_zero(ibv_destroy_flow(rxq->mac_flow[mac_index][vlan_index])); + rxq->mac_flow[mac_index][vlan_index] = NULL; +} + +/** + * Unregister a MAC address from a RX queue. + * + * @param rxq + * Pointer to RX queue structure. + * @param mac_index + * MAC address index. + */ +static void +rxq_mac_addr_del(struct rxq *rxq, unsigned int mac_index) +{ + struct priv *priv = rxq->priv; + unsigned int i; + unsigned int vlans = 0; assert(mac_index < elemof(priv->mac)); - if (!BITFIELD_ISSET(rxq->mac_configured, mac_index)) { - assert(rxq->mac_flow[mac_index] == NULL); + if (!BITFIELD_ISSET(rxq->mac_configured, mac_index)) return; + for (i = 0; (i != elemof(priv->vlan_filter)); ++i) { + if (!priv->vlan_filter[i].enabled) + continue; + rxq_del_flow(rxq, mac_index, i); + vlans++; + } + if (!vlans) { + rxq_del_flow(rxq, mac_index, 0); } - DEBUG("%p: removing MAC address %02x:%02x:%02x:%02x:%02x:%02x" - " index %u", - (void *)rxq, - (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5], - mac_index); - assert(rxq->mac_flow[mac_index] != NULL); - claim_zero(ibv_exp_destroy_flow(rxq->mac_flow[mac_index])); - rxq->mac_flow[mac_index] = NULL; BITFIELD_RESET(rxq->mac_configured, mac_index); } @@ -1868,111 +2488,76 @@ static int rxq_promiscuous_enable(struct rxq *); static void rxq_promiscuous_disable(struct rxq *); /** - * Register a MAC address in a RX queue. + * Add single flow steering rule. * * @param rxq * Pointer to RX queue structure. * @param mac_index * MAC address index to register. + * @param vlan_index + * VLAN index. Use -1 for a flow without VLAN. * * @return * 0 on success, errno value on failure. */ static int -rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index) +rxq_add_flow(struct rxq *rxq, unsigned int mac_index, unsigned int vlan_index) { + struct ibv_flow *flow; struct priv *priv = rxq->priv; const uint8_t (*mac)[ETHER_ADDR_LEN] = - (const uint8_t (*)[ETHER_ADDR_LEN]) - priv->mac[mac_index].addr_bytes; - unsigned int vlans = 0; - unsigned int specs = 0; - unsigned int i, j; - struct ibv_exp_flow *flow; - - assert(mac_index < elemof(priv->mac)); - if (BITFIELD_ISSET(rxq->mac_configured, mac_index)) - rxq_mac_addr_del(rxq, mac_index); - /* Number of configured VLANs. */ - for (i = 0; (i != elemof(priv->vlan_filter)); ++i) - if (priv->vlan_filter[i].enabled) - ++vlans; - specs = (vlans ? vlans : 1); + (const uint8_t (*)[ETHER_ADDR_LEN]) + priv->mac[mac_index].addr_bytes; /* Allocate flow specification on the stack. */ - struct ibv_exp_flow_attr data - [1 + - (sizeof(struct ibv_exp_flow_spec_eth[specs]) / - sizeof(struct ibv_exp_flow_attr)) + - !!(sizeof(struct ibv_exp_flow_spec_eth[specs]) % - sizeof(struct ibv_exp_flow_attr))]; - struct ibv_exp_flow_attr *attr = (void *)&data[0]; - struct ibv_exp_flow_spec_eth *spec = (void *)&data[1]; + struct __attribute__((packed)) { + struct ibv_flow_attr attr; + struct ibv_flow_spec_eth spec; + } data; + struct ibv_flow_attr *attr = &data.attr; + struct ibv_flow_spec_eth *spec = &data.spec; + assert(mac_index < elemof(priv->mac)); + assert((vlan_index < elemof(priv->vlan_filter)) || (vlan_index == -1u)); /* * No padding must be inserted by the compiler between attr and spec. * This layout is expected by libibverbs. */ assert(((uint8_t *)attr + sizeof(*attr)) == (uint8_t *)spec); - *attr = (struct ibv_exp_flow_attr){ - .type = IBV_EXP_FLOW_ATTR_NORMAL, - .num_of_specs = specs, + *attr = (struct ibv_flow_attr){ + .type = IBV_FLOW_ATTR_NORMAL, + .num_of_specs = 1, .port = priv->port, .flags = 0 }; - *spec = (struct ibv_exp_flow_spec_eth){ - .type = IBV_EXP_FLOW_SPEC_ETH, + *spec = (struct ibv_flow_spec_eth){ + .type = IBV_FLOW_SPEC_ETH, .size = sizeof(*spec), .val = { .dst_mac = { (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5] - } + }, + .vlan_tag = ((vlan_index != -1u) ? + htons(priv->vlan_filter[vlan_index].id) : + 0), }, .mask = { .dst_mac = "\xff\xff\xff\xff\xff\xff", - .vlan_tag = (vlans ? htons(0xfff) : 0) + .vlan_tag = ((vlan_index != -1u) ? htons(0xfff) : 0), } }; - /* Fill VLAN specifications. */ - for (i = 0, j = 0; (i != elemof(priv->vlan_filter)); ++i) { - if (!priv->vlan_filter[i].enabled) - continue; - assert(j != vlans); - if (j) - spec[j] = spec[0]; - spec[j].val.vlan_tag = htons(priv->vlan_filter[i].id); - ++j; - } DEBUG("%p: adding MAC address %02x:%02x:%02x:%02x:%02x:%02x index %u" - " (%u VLAN(s) configured)", + " (VLAN %s %" PRIu16 ")", (void *)rxq, (*mac)[0], (*mac)[1], (*mac)[2], (*mac)[3], (*mac)[4], (*mac)[5], mac_index, - vlans); + ((vlan_index != -1u) ? "ID" : "index"), + ((vlan_index != -1u) ? priv->vlan_filter[vlan_index].id : -1u)); /* Create related flow. */ errno = 0; - flow = ibv_exp_create_flow(rxq->qp, attr); + flow = ibv_create_flow(rxq->qp, attr); if (flow == NULL) { - int err = errno; - - /* Flow creation failure is not fatal when in DMFS A0 mode. - * Ignore error if promiscuity is already enabled or can be - * enabled. */ - if (priv->promisc_ok) - return 0; - if ((rxq->promisc_flow != NULL) || - (rxq_promiscuous_enable(rxq) == 0)) { - if (rxq->promisc_flow != NULL) - rxq_promiscuous_disable(rxq); - WARN("cannot configure normal flow but promiscuous" - " mode is fine, assuming promiscuous optimization" - " is enabled" - " (options mlx4_core log_num_mgm_entry_size=-7)"); - priv->promisc_ok = 1; - return 0; - } - errno = err; /* It's not clear whether errno is always set in this case. */ ERROR("%p: flow configuration failed, errno=%d: %s", (void *)rxq, errno, @@ -1981,8 +2566,58 @@ rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index) return errno; return EINVAL; } - assert(rxq->mac_flow[mac_index] == NULL); - rxq->mac_flow[mac_index] = flow; + if (vlan_index == -1u) + vlan_index = 0; + assert(rxq->mac_flow[mac_index][vlan_index] == NULL); + rxq->mac_flow[mac_index][vlan_index] = flow; + return 0; +} + +/** + * Register a MAC address in a RX queue. + * + * @param rxq + * Pointer to RX queue structure. + * @param mac_index + * MAC address index to register. + * + * @return + * 0 on success, errno value on failure. + */ +static int +rxq_mac_addr_add(struct rxq *rxq, unsigned int mac_index) +{ + struct priv *priv = rxq->priv; + unsigned int i; + unsigned int vlans = 0; + int ret; + + assert(mac_index < elemof(priv->mac)); + if (BITFIELD_ISSET(rxq->mac_configured, mac_index)) + rxq_mac_addr_del(rxq, mac_index); + /* Fill VLAN specifications. */ + for (i = 0; (i != elemof(priv->vlan_filter)); ++i) { + if (!priv->vlan_filter[i].enabled) + continue; + /* Create related flow. */ + ret = rxq_add_flow(rxq, mac_index, i); + if (!ret) { + vlans++; + continue; + } + /* Failure, rollback. */ + while (i != 0) + if (priv->vlan_filter[--i].enabled) + rxq_del_flow(rxq, mac_index, i); + assert(ret > 0); + return ret; + } + /* In case there is no VLAN filter. */ + if (!vlans) { + ret = rxq_add_flow(rxq, mac_index, -1); + if (ret) + return ret; + } BITFIELD_SET(rxq->mac_configured, mac_index); return 0; } @@ -2139,9 +2774,9 @@ end: static int rxq_allmulticast_enable(struct rxq *rxq) { - struct ibv_exp_flow *flow; - struct ibv_exp_flow_attr attr = { - .type = IBV_EXP_FLOW_ATTR_MC_DEFAULT, + struct ibv_flow *flow; + struct ibv_flow_attr attr = { + .type = IBV_FLOW_ATTR_MC_DEFAULT, .num_of_specs = 0, .port = rxq->priv->port, .flags = 0 @@ -2151,7 +2786,7 @@ rxq_allmulticast_enable(struct rxq *rxq) if (rxq->allmulti_flow != NULL) return EBUSY; errno = 0; - flow = ibv_exp_create_flow(rxq->qp, &attr); + flow = ibv_create_flow(rxq->qp, &attr); if (flow == NULL) { /* It's not clear whether errno is always set in this case. */ ERROR("%p: flow configuration failed, errno=%d: %s", @@ -2178,7 +2813,7 @@ rxq_allmulticast_disable(struct rxq *rxq) DEBUG("%p: disabling allmulticast mode", (void *)rxq); if (rxq->allmulti_flow == NULL) return; - claim_zero(ibv_exp_destroy_flow(rxq->allmulti_flow)); + claim_zero(ibv_destroy_flow(rxq->allmulti_flow)); rxq->allmulti_flow = NULL; DEBUG("%p: allmulticast mode disabled", (void *)rxq); } @@ -2195,9 +2830,9 @@ rxq_allmulticast_disable(struct rxq *rxq) static int rxq_promiscuous_enable(struct rxq *rxq) { - struct ibv_exp_flow *flow; - struct ibv_exp_flow_attr attr = { - .type = IBV_EXP_FLOW_ATTR_ALL_DEFAULT, + struct ibv_flow *flow; + struct ibv_flow_attr attr = { + .type = IBV_FLOW_ATTR_ALL_DEFAULT, .num_of_specs = 0, .port = rxq->priv->port, .flags = 0 @@ -2209,7 +2844,7 @@ rxq_promiscuous_enable(struct rxq *rxq) if (rxq->promisc_flow != NULL) return EBUSY; errno = 0; - flow = ibv_exp_create_flow(rxq->qp, &attr); + flow = ibv_create_flow(rxq->qp, &attr); if (flow == NULL) { /* It's not clear whether errno is always set in this case. */ ERROR("%p: flow configuration failed, errno=%d: %s", @@ -2238,7 +2873,7 @@ rxq_promiscuous_disable(struct rxq *rxq) DEBUG("%p: disabling promiscuous mode", (void *)rxq); if (rxq->promisc_flow == NULL) return; - claim_zero(ibv_exp_destroy_flow(rxq->promisc_flow)); + claim_zero(ibv_destroy_flow(rxq->promisc_flow)); rxq->promisc_flow = NULL; DEBUG("%p: promiscuous mode disabled", (void *)rxq); } @@ -2254,11 +2889,35 @@ rxq_promiscuous_disable(struct rxq *rxq) static void rxq_cleanup(struct rxq *rxq) { + struct ibv_exp_release_intf_params params; + DEBUG("cleaning up %p", (void *)rxq); if (rxq->sp) rxq_free_elts_sp(rxq); else rxq_free_elts(rxq); + if (rxq->if_qp != NULL) { + assert(rxq->priv != NULL); + assert(rxq->priv->ctx != NULL); + assert(rxq->qp != NULL); + params = (struct ibv_exp_release_intf_params){ + .comp_mask = 0, + }; + claim_zero(ibv_exp_release_intf(rxq->priv->ctx, + rxq->if_qp, + ¶ms)); + } + if (rxq->if_cq != NULL) { + assert(rxq->priv != NULL); + assert(rxq->priv->ctx != NULL); + assert(rxq->cq != NULL); + params = (struct ibv_exp_release_intf_params){ + .comp_mask = 0, + }; + claim_zero(ibv_exp_release_intf(rxq->priv->ctx, + rxq->if_cq, + ¶ms)); + } if (rxq->qp != NULL) { rxq_promiscuous_disable(rxq); rxq_allmulticast_disable(rxq); @@ -2267,11 +2926,97 @@ rxq_cleanup(struct rxq *rxq) } if (rxq->cq != NULL) claim_zero(ibv_destroy_cq(rxq->cq)); + if (rxq->rd != NULL) { + struct ibv_exp_destroy_res_domain_attr attr = { + .comp_mask = 0, + }; + + assert(rxq->priv != NULL); + assert(rxq->priv->ctx != NULL); + claim_zero(ibv_exp_destroy_res_domain(rxq->priv->ctx, + rxq->rd, + &attr)); + } if (rxq->mr != NULL) claim_zero(ibv_dereg_mr(rxq->mr)); memset(rxq, 0, sizeof(*rxq)); } +/** + * Translate RX completion flags to packet type. + * + * @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. + */ +static inline uint32_t +rxq_cq_to_pkt_type(uint32_t flags) +{ + uint32_t pkt_type; + + if (flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) + pkt_type = + TRANSPOSE(flags, + IBV_EXP_CQ_RX_OUTER_IPV4_PACKET, RTE_PTYPE_L3_IPV4) | + TRANSPOSE(flags, + IBV_EXP_CQ_RX_OUTER_IPV6_PACKET, RTE_PTYPE_L3_IPV6) | + TRANSPOSE(flags, + IBV_EXP_CQ_RX_IPV4_PACKET, RTE_PTYPE_INNER_L3_IPV4) | + TRANSPOSE(flags, + IBV_EXP_CQ_RX_IPV6_PACKET, RTE_PTYPE_INNER_L3_IPV6); + else + pkt_type = + TRANSPOSE(flags, + IBV_EXP_CQ_RX_IPV4_PACKET, RTE_PTYPE_L3_IPV4) | + TRANSPOSE(flags, + IBV_EXP_CQ_RX_IPV6_PACKET, RTE_PTYPE_L3_IPV6); + return pkt_type; +} + +/** + * Translate RX completion flags to offload flags. + * + * @param[in] rxq + * Pointer to RX queue structure. + * @param flags + * RX completion flags returned by poll_length_flags(). + * + * @return + * Offload flags (ol_flags) for struct rte_mbuf. + */ +static inline uint32_t +rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags) +{ + uint32_t ol_flags = 0; + + if (rxq->csum) + ol_flags |= + TRANSPOSE(~flags, + IBV_EXP_CQ_RX_IP_CSUM_OK, + PKT_RX_IP_CKSUM_BAD) | + 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). + */ + if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun)) + ol_flags |= + TRANSPOSE(~flags, + IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK, + PKT_RX_IP_CKSUM_BAD) | + TRANSPOSE(~flags, + IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK, + PKT_RX_L4_CKSUM_BAD); + return ol_flags; +} + static uint16_t mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n); @@ -2293,59 +3038,82 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) { struct rxq *rxq = (struct rxq *)dpdk_rxq; struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp; - struct ibv_wc wcs[pkts_n]; + const unsigned int elts_n = rxq->elts_n; + unsigned int elts_head = rxq->elts_head; struct ibv_recv_wr head; struct ibv_recv_wr **next = &head.next; struct ibv_recv_wr *bad_wr; - int ret = 0; - int wcs_n; - int i; + unsigned int i; + unsigned int pkts_ret = 0; + int ret; if (unlikely(!rxq->sp)) return mlx4_rx_burst(dpdk_rxq, pkts, pkts_n); if (unlikely(elts == NULL)) /* See RTE_DEV_CMD_SET_MTU. */ return 0; - wcs_n = ibv_poll_cq(rxq->cq, pkts_n, wcs); - if (unlikely(wcs_n == 0)) - return 0; - if (unlikely(wcs_n < 0)) { - DEBUG("rxq=%p, ibv_poll_cq() failed (wc_n=%d)", - (void *)rxq, wcs_n); - return 0; - } - assert(wcs_n <= (int)pkts_n); - /* For each work completion. */ - for (i = 0; (i != wcs_n); ++i) { - struct ibv_wc *wc = &wcs[i]; - uint64_t wr_id = wc->wr_id; - uint32_t len = wc->byte_len; - struct rxq_elt_sp *elt = &(*elts)[wr_id]; + for (i = 0; (i != pkts_n); ++i) { + struct rxq_elt_sp *elt = &(*elts)[elts_head]; struct ibv_recv_wr *wr = &elt->wr; + uint64_t wr_id = wr->wr_id; + unsigned int len; + unsigned int pkt_buf_len; struct rte_mbuf *pkt_buf = NULL; /* Buffer returned in pkts. */ struct rte_mbuf **pkt_buf_next = &pkt_buf; unsigned int seg_headroom = RTE_PKTMBUF_HEADROOM; unsigned int j = 0; + uint32_t flags; /* Sanity checks. */ +#ifdef NDEBUG + (void)wr_id; +#endif assert(wr_id < rxq->elts_n); - assert(wr_id == wr->wr_id); assert(wr->sg_list == elt->sges); assert(wr->num_sge == elemof(elt->sges)); - /* Link completed WRs together for repost. */ - *next = wr; - next = &wr->next; - if (unlikely(wc->status != IBV_WC_SUCCESS)) { - /* Whatever, just repost the offending WR. */ - DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work completion" - " status (%d): %s", - (void *)rxq, wc->wr_id, wc->status, - ibv_wc_status_str(wc->status)); + assert(elts_head < rxq->elts_n); + assert(rxq->elts_head < rxq->elts_n); + ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL, + &flags); + if (unlikely(ret < 0)) { + struct ibv_wc wc; + int wcs_n; + + DEBUG("rxq=%p, poll_length() failed (ret=%d)", + (void *)rxq, ret); + /* ibv_poll_cq() must be used in case of failure. */ + wcs_n = ibv_poll_cq(rxq->cq, 1, &wc); + if (unlikely(wcs_n == 0)) + break; + if (unlikely(wcs_n < 0)) { + DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)", + (void *)rxq, wcs_n); + break; + } + assert(wcs_n == 1); + if (unlikely(wc.status != IBV_WC_SUCCESS)) { + /* Whatever, just repost the offending WR. */ + DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work" + " completion status (%d): %s", + (void *)rxq, wc.wr_id, wc.status, + ibv_wc_status_str(wc.status)); #ifdef MLX4_PMD_SOFT_COUNTERS - /* Increase dropped packets counter. */ - ++rxq->stats.idropped; + /* Increment dropped packets counter. */ + ++rxq->stats.idropped; #endif - goto repost; + /* Link completed WRs together for repost. */ + *next = wr; + next = &wr->next; + goto repost; + } + ret = wc.byte_len; } + if (ret == 0) + break; + len = ret; + pkt_buf_len = len; + /* Link completed WRs together for repost. */ + *next = wr; + next = &wr->next; /* * Replace spent segments with new ones, concatenate and * return them as pkt_buf. @@ -2361,7 +3129,7 @@ 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, @@ -2370,8 +3138,10 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) DEBUG("rxq=%p, wr_id=%" PRIu64 ":" " can't allocate a new mbuf", (void *)rxq, wr_id); - if (pkt_buf != NULL) + if (pkt_buf != NULL) { + *pkt_buf_next = NULL; rte_pktmbuf_free(pkt_buf); + } /* Increase out of memory counters. */ ++rxq->stats.rx_nombuf; ++rxq->priv->dev->data->rx_mbuf_alloc_failed; @@ -2388,7 +3158,6 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) rep->ol_flags = -1; #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); @@ -2428,39 +3197,44 @@ mlx4_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) assert(j != 0); NB_SEGS(pkt_buf) = j; PORT(pkt_buf) = rxq->port_id; - PKT_LEN(pkt_buf) = wc->byte_len; - pkt_buf->ol_flags = 0; + 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; - ++ret; + ++pkts_ret; #ifdef MLX4_PMD_SOFT_COUNTERS /* Increase bytes counter. */ - rxq->stats.ibytes += wc->byte_len; + rxq->stats.ibytes += pkt_buf_len; #endif repost: + if (++elts_head >= elts_n) + elts_head = 0; continue; } + if (unlikely(i == 0)) + return 0; *next = NULL; /* Repost WRs. */ #ifdef DEBUG_RECV - DEBUG("%p: reposting %d WRs starting from %" PRIu64 " (%p)", - (void *)rxq, wcs_n, wcs[0].wr_id, (void *)head.next); + DEBUG("%p: reposting %d WRs", (void *)rxq, i); #endif - i = ibv_post_recv(rxq->qp, head.next, &bad_wr); - if (unlikely(i)) { + ret = ibv_post_recv(rxq->qp, head.next, &bad_wr); + if (unlikely(ret)) { /* Inability to repost WRs is fatal. */ DEBUG("%p: ibv_post_recv(): failed for WR %p: %s", (void *)rxq->priv, (void *)bad_wr, - strerror(i)); + strerror(ret)); abort(); } + rxq->elts_head = elts_head; #ifdef MLX4_PMD_SOFT_COUNTERS /* Increase packets counter. */ - rxq->stats.ipackets += ret; + rxq->stats.ipackets += pkts_ret; #endif - return ret; + return pkts_ret; } /** @@ -2485,62 +3259,75 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) { struct rxq *rxq = (struct rxq *)dpdk_rxq; struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp; - struct ibv_wc wcs[pkts_n]; - struct ibv_recv_wr head; - struct ibv_recv_wr **next = &head.next; - struct ibv_recv_wr *bad_wr; - int ret = 0; - int wcs_n; - int i; + const unsigned int elts_n = rxq->elts_n; + unsigned int elts_head = rxq->elts_head; + struct ibv_sge sges[pkts_n]; + unsigned int i; + unsigned int pkts_ret = 0; + int ret; if (unlikely(rxq->sp)) return mlx4_rx_burst_sp(dpdk_rxq, pkts, pkts_n); - wcs_n = ibv_poll_cq(rxq->cq, pkts_n, wcs); - if (unlikely(wcs_n == 0)) - return 0; - if (unlikely(wcs_n < 0)) { - DEBUG("rxq=%p, ibv_poll_cq() failed (wc_n=%d)", - (void *)rxq, wcs_n); - return 0; - } - assert(wcs_n <= (int)pkts_n); - /* For each work completion. */ - for (i = 0; (i != wcs_n); ++i) { - struct ibv_wc *wc = &wcs[i]; - uint64_t wr_id = wc->wr_id; - uint32_t len = wc->byte_len; - struct rxq_elt *elt = &(*elts)[WR_ID(wr_id).id]; + for (i = 0; (i != pkts_n); ++i) { + struct rxq_elt *elt = &(*elts)[elts_head]; struct ibv_recv_wr *wr = &elt->wr; + uint64_t wr_id = wr->wr_id; + unsigned int len; struct rte_mbuf *seg = (void *)((uintptr_t)elt->sge.addr - WR_ID(wr_id).offset); struct rte_mbuf *rep; + uint32_t flags; /* Sanity checks. */ assert(WR_ID(wr_id).id < rxq->elts_n); - assert(wr_id == wr->wr_id); assert(wr->sg_list == &elt->sge); assert(wr->num_sge == 1); - /* Link completed WRs together for repost. */ - *next = wr; - next = &wr->next; - if (unlikely(wc->status != IBV_WC_SUCCESS)) { - /* Whatever, just repost the offending WR. */ - DEBUG("rxq=%p, wr_id=%" PRIu32 ": bad work completion" - " status (%d): %s", - (void *)rxq, WR_ID(wr_id).id, wc->status, - ibv_wc_status_str(wc->status)); -#ifdef MLX4_PMD_SOFT_COUNTERS - /* Increase dropped packets counter. */ - ++rxq->stats.idropped; -#endif - goto repost; - } + 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_prefetch0(seg); - rep = __rte_mbuf_raw_alloc(rxq->mp); + 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)) { + struct ibv_wc wc; + int wcs_n; + + DEBUG("rxq=%p, poll_length() failed (ret=%d)", + (void *)rxq, ret); + /* ibv_poll_cq() must be used in case of failure. */ + wcs_n = ibv_poll_cq(rxq->cq, 1, &wc); + if (unlikely(wcs_n == 0)) + break; + if (unlikely(wcs_n < 0)) { + DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)", + (void *)rxq, wcs_n); + break; + } + assert(wcs_n == 1); + if (unlikely(wc.status != IBV_WC_SUCCESS)) { + /* Whatever, just repost the offending WR. */ + DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work" + " 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; + } + ret = wc.byte_len; + } + if (ret == 0) + break; + len = ret; + rep = rte_mbuf_raw_alloc(rxq->mp); if (unlikely(rep == NULL)) { /* * Unable to allocate a replacement mbuf, @@ -2563,6 +3350,9 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) (uintptr_t)rep); assert(WR_ID(wr->wr_id).id == WR_ID(wr_id).id); + /* Add SGE to array for repost. */ + sges[i] = elt->sge; + /* Update seg information. */ SET_DATA_OFF(seg, RTE_PKTMBUF_HEADROOM); NB_SEGS(seg) = 1; @@ -2570,89 +3360,86 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) NEXT(seg) = NULL; PKT_LEN(seg) = len; DATA_LEN(seg) = len; - seg->ol_flags = 0; + seg->packet_type = rxq_cq_to_pkt_type(flags); + seg->ol_flags = rxq_cq_to_ol_flags(rxq, flags); /* Return packet. */ *(pkts++) = seg; - ++ret; + ++pkts_ret; #ifdef MLX4_PMD_SOFT_COUNTERS /* Increase bytes counter. */ - rxq->stats.ibytes += wc->byte_len; + rxq->stats.ibytes += len; #endif repost: + if (++elts_head >= elts_n) + elts_head = 0; continue; } - *next = NULL; + if (unlikely(i == 0)) + return 0; /* Repost WRs. */ #ifdef DEBUG_RECV - DEBUG("%p: reposting %d WRs starting from %" PRIu32 " (%p)", - (void *)rxq, wcs_n, WR_ID(wcs[0].wr_id).id, (void *)head.next); + DEBUG("%p: reposting %u WRs", (void *)rxq, i); #endif - i = ibv_post_recv(rxq->qp, head.next, &bad_wr); - if (unlikely(i)) { + ret = rxq->if_qp->recv_burst(rxq->qp, sges, i); + if (unlikely(ret)) { /* Inability to repost WRs is fatal. */ - DEBUG("%p: ibv_post_recv(): failed for WR %p: %s", + DEBUG("%p: recv_burst(): failed (ret=%d)", (void *)rxq->priv, - (void *)bad_wr, - strerror(i)); + ret); abort(); } + rxq->elts_head = elts_head; #ifdef MLX4_PMD_SOFT_COUNTERS /* Increase packets counter. */ - rxq->stats.ipackets += ret; + rxq->stats.ipackets += pkts_ret; #endif - return ret; + return pkts_ret; } -#ifdef INLINE_RECV - /** - * Allocate a Queue Pair in case inline receive is supported. + * DPDK callback for RX in secondary processes. * - * @param priv - * Pointer to private structure. - * @param cq - * Completion queue to associate with QP. - * @param desc - * Number of descriptors in QP (hint only). + * This function configures all queues from primary process information + * if necessary before reverting to the normal RX burst callback. + * + * @param dpdk_rxq + * Generic pointer to RX queue structure. + * @param[out] pkts + * Array to store received packets. + * @param pkts_n + * Maximum number of packets in array. * * @return - * QP pointer or NULL in case of error. - */ -static struct ibv_qp * -rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) -{ - struct ibv_exp_qp_init_attr attr = { - /* CQ to be associated with the send queue. */ - .send_cq = cq, - /* CQ to be associated with the receive queue. */ - .recv_cq = cq, - .max_inl_recv = priv->inl_recv_size, - .cap = { - /* Max number of outstanding WRs. */ - .max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ? - priv->device_attr.max_qp_wr : - desc), - /* Max number of scatter/gather elements in a WR. */ - .max_recv_sge = ((priv->device_attr.max_sge < - MLX4_PMD_SGE_WR_N) ? - priv->device_attr.max_sge : - MLX4_PMD_SGE_WR_N), - }, - .qp_type = IBV_QPT_RAW_PACKET, - .pd = priv->pd - }; - - attr.comp_mask = IBV_EXP_QP_INIT_ATTR_PD; - attr.comp_mask |= IBV_EXP_QP_INIT_ATTR_INL_RECV; + * Number of packets successfully received (<= pkts_n). + */ +static uint16_t +mlx4_rx_burst_secondary_setup(void *dpdk_rxq, struct rte_mbuf **pkts, + uint16_t pkts_n) +{ + struct rxq *rxq = dpdk_rxq; + struct priv *priv = mlx4_secondary_data_setup(rxq->priv); + struct priv *primary_priv; + unsigned int index; - return ibv_exp_create_qp(priv->ctx, &attr); + if (priv == NULL) + return 0; + primary_priv = + mlx4_secondary_data[priv->dev->data->port_id].primary_priv; + /* Look for queue index in both private structures. */ + for (index = 0; index != priv->rxqs_n; ++index) + if (((*primary_priv->rxqs)[index] == rxq) || + ((*priv->rxqs)[index] == rxq)) + break; + if (index == priv->rxqs_n) + return 0; + rxq = (*priv->rxqs)[index]; + return priv->dev->rx_pkt_burst(rxq, pkts, pkts_n); } -#else /* INLINE_RECV */ - /** * Allocate a Queue Pair. + * Optionally setup inline receive if supported. * * @param priv * Pointer to private structure. @@ -2665,9 +3452,10 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) * QP pointer or NULL in case of error. */ static struct ibv_qp * -rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) +rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc, + struct ibv_exp_res_domain *rd) { - struct ibv_qp_init_attr attr = { + struct ibv_exp_qp_init_attr attr = { /* CQ to be associated with the send queue. */ .send_cq = cq, /* CQ to be associated with the receive queue. */ @@ -2683,18 +3471,25 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) priv->device_attr.max_sge : MLX4_PMD_SGE_WR_N), }, - .qp_type = IBV_QPT_RAW_PACKET + .qp_type = IBV_QPT_RAW_PACKET, + .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD | + IBV_EXP_QP_INIT_ATTR_RES_DOMAIN), + .pd = priv->pd, + .res_domain = rd, }; - return ibv_create_qp(priv->pd, &attr); +#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); } -#endif /* INLINE_RECV */ - #ifdef RSS_SUPPORT /** * Allocate a RSS Queue Pair. + * Optionally setup inline receive if supported. * * @param priv * Pointer to private structure. @@ -2710,16 +3505,13 @@ rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc) */ static struct ibv_qp * rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc, - int parent) + int parent, struct ibv_exp_res_domain *rd) { struct ibv_exp_qp_init_attr attr = { /* CQ to be associated with the send queue. */ .send_cq = cq, /* CQ to be associated with the receive queue. */ .recv_cq = cq, -#ifdef INLINE_RECV - .max_inl_recv = priv->inl_recv_size, -#endif .cap = { /* Max number of outstanding WRs. */ .max_recv_wr = ((priv->device_attr.max_qp_wr < desc) ? @@ -2733,18 +3525,22 @@ rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc, }, .qp_type = IBV_QPT_RAW_PACKET, .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD | + IBV_EXP_QP_INIT_ATTR_RES_DOMAIN | IBV_EXP_QP_INIT_ATTR_QPG), - .pd = priv->pd + .pd = priv->pd, + .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) { 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(priv->rxqs_n + 1) >> 1; DEBUG("initializing parent RSS queue"); } else { attr.qpg.qpg_type = IBV_EXP_QPG_CHILD_RX; @@ -2782,6 +3578,7 @@ 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); @@ -2790,14 +3587,25 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) (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)); mbuf_n = desc_n; + /* Toggle RX checksum offload if hardware supports it. */ + if (priv->hw_csum) { + tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum; + rxq->csum = tmpl.csum; + } + if (priv->hw_csum_l2tun) { + tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum; + rxq->csum_l2tun = tmpl.csum_l2tun; + } /* Enable scattered packets support for this queue if necessary. */ + assert(mb_len >= RTE_PKTMBUF_HEADROOM); if ((dev->data->dev_conf.rxmode.jumbo_frame) && (dev->data->dev_conf.rxmode.max_rx_pkt_len > - (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) { + (mb_len - RTE_PKTMBUF_HEADROOM))) { tmpl.sp = 1; desc_n /= MLX4_PMD_SGE_WR_N; } else @@ -2958,6 +3766,9 @@ skip_rtr: * 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 @@ -2968,7 +3779,7 @@ skip_rtr: */ static int rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, - unsigned int socket, const struct rte_eth_rxconf *conf, + unsigned int socket, int inactive, const struct rte_eth_rxconf *conf, struct rte_mempool *mp) { struct priv *priv = dev->data->dev_private; @@ -2978,8 +3789,14 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, .socket = socket }; struct ibv_exp_qp_attr mod; + union { + struct ibv_exp_query_intf_params params; + struct ibv_exp_cq_init_attr cq; + struct ibv_exp_res_domain_init_attr rd; + } attr; + enum ibv_exp_query_intf_status status; struct ibv_recv_wr *bad_wr; - struct rte_mbuf *buf; + unsigned int mb_len; int ret = 0; int parent = (rxq == &priv->rxq_parent); @@ -2995,37 +3812,29 @@ 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, desc); + " 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. */ + assert(mb_len >= RTE_PKTMBUF_HEADROOM); if ((dev->data->dev_conf.rxmode.jumbo_frame) && (dev->data->dev_conf.rxmode.max_rx_pkt_len > - (tmpl.mb_len - RTE_PKTMBUF_HEADROOM))) { + (mb_len - RTE_PKTMBUF_HEADROOM))) { tmpl.sp = 1; desc /= MLX4_PMD_SGE_WR_N; } 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", @@ -3033,7 +3842,24 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, goto error; } skip_mr: - tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, NULL, 0); + attr.rd = (struct ibv_exp_res_domain_init_attr){ + .comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL | + IBV_EXP_RES_DOMAIN_MSG_MODEL), + .thread_model = IBV_EXP_THREAD_SINGLE, + .msg_model = IBV_EXP_MSG_HIGH_BW, + }; + tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd); + if (tmpl.rd == NULL) { + ret = ENOMEM; + ERROR("%p: RD 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); if (tmpl.cq == NULL) { ret = ENOMEM; ERROR("%p: CQ creation failure: %s", @@ -3045,11 +3871,12 @@ skip_mr: 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); + if (priv->rss && !inactive) + tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent, + tmpl.rd); else #endif /* RSS_SUPPORT */ - tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc); + tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd); if (tmpl.qp == NULL) { ret = (errno ? errno : EINVAL); ERROR("%p: QP creation failure: %s", @@ -3119,6 +3946,28 @@ skip_alloc: /* Save port ID. */ tmpl.port_id = dev->data->port_id; DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id); + attr.params = (struct ibv_exp_query_intf_params){ + .intf_scope = IBV_EXP_INTF_GLOBAL, + .intf = IBV_EXP_INTF_CQ, + .obj = tmpl.cq, + }; + tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status); + if (tmpl.if_cq == NULL) { + 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); @@ -3158,8 +4007,11 @@ mlx4_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, { struct priv *priv = dev->data->dev_private; struct rxq *rxq = (*priv->rxqs)[idx]; + int inactive = 0; int ret; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; priv_lock(priv); DEBUG("%p: configuring queue %u for %u descriptors", (void *)dev, idx, desc); @@ -3187,7 +4039,9 @@ 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 (idx >= rte_align32pow2(priv->rxqs_n + 1) >> 1) + inactive = 1; + ret = rxq_setup(dev, rxq, desc, socket, inactive, conf, mp); if (ret) rte_free(rxq); else { @@ -3218,6 +4072,8 @@ mlx4_rx_queue_release(void *dpdk_rxq) struct priv *priv; unsigned int i; + if (mlx4_is_secondary()) + return; if (rxq == NULL) return; priv = rxq->priv; @@ -3235,6 +4091,9 @@ mlx4_rx_queue_release(void *dpdk_rxq) priv_unlock(priv); } +static void +priv_dev_interrupt_handler_install(struct priv *, struct rte_eth_dev *); + /** * DPDK callback to start the device. * @@ -3254,6 +4113,8 @@ mlx4_dev_start(struct rte_eth_dev *dev) unsigned int r; struct rxq *rxq; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; priv_lock(priv); if (priv->started) { priv_unlock(priv); @@ -3294,8 +4155,10 @@ mlx4_dev_start(struct rte_eth_dev *dev) } } priv->started = 0; + priv_unlock(priv); return -ret; } while ((--r) && ((rxq = (*priv->rxqs)[++i]), i)); + priv_dev_interrupt_handler_install(priv, dev); priv_unlock(priv); return 0; } @@ -3316,6 +4179,8 @@ mlx4_dev_stop(struct rte_eth_dev *dev) unsigned int r; struct rxq *rxq; + if (mlx4_is_secondary()) + return; priv_lock(priv); if (!priv->started) { priv_unlock(priv); @@ -3392,6 +4257,9 @@ removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) return 0; } +static void +priv_dev_interrupt_handler_uninstall(struct priv *, struct rte_eth_dev *); + /** * DPDK callback to close the device. * @@ -3403,10 +4271,12 @@ removed_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) static void mlx4_dev_close(struct rte_eth_dev *dev) { - struct priv *priv = dev->data->dev_private; + struct priv *priv = mlx4_get_priv(dev); void *tmp; unsigned int i; + if (priv == NULL) + return; priv_lock(priv); DEBUG("%p: closing device \"%s\"", (void *)dev, @@ -3452,10 +4322,95 @@ mlx4_dev_close(struct rte_eth_dev *dev) claim_zero(ibv_close_device(priv->ctx)); } else assert(priv->ctx == NULL); + priv_dev_interrupt_handler_uninstall(priv, dev); 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. * @@ -3467,9 +4422,12 @@ mlx4_dev_close(struct rte_eth_dev *dev) static void mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) { - struct priv *priv = dev->data->dev_private; + struct priv *priv = mlx4_get_priv(dev); unsigned int max; + char ifname[IF_NAMESIZE]; + if (priv == NULL) + return; priv_lock(priv); /* FIXME: we should ask the device for these values. */ info->min_rx_bufsize = 32; @@ -3485,10 +4443,49 @@ 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 | + DEV_RX_OFFLOAD_UDP_CKSUM | + DEV_RX_OFFLOAD_TCP_CKSUM) : + 0); + info->tx_offload_capa = + (priv->hw_csum ? + (DEV_TX_OFFLOAD_IPV4_CKSUM | + DEV_TX_OFFLOAD_UDP_CKSUM | + DEV_TX_OFFLOAD_TCP_CKSUM) : + 0); + 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); } +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. * @@ -3500,11 +4497,13 @@ mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) static void mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) { - struct priv *priv = dev->data->dev_private; + struct priv *priv = mlx4_get_priv(dev); struct rte_eth_stats tmp = {0}; unsigned int i; unsigned int idx; + if (priv == NULL) + return; priv_lock(priv); /* Add software counters. */ for (i = 0; (i != priv->rxqs_n); ++i) { @@ -3563,10 +4562,12 @@ mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) static void mlx4_stats_reset(struct rte_eth_dev *dev) { - struct priv *priv = dev->data->dev_private; + struct priv *priv = mlx4_get_priv(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) @@ -3578,7 +4579,7 @@ 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 }; } @@ -3601,14 +4602,13 @@ mlx4_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index) { struct priv *priv = dev->data->dev_private; + if (mlx4_is_secondary()) + return; priv_lock(priv); 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: @@ -3633,15 +4633,14 @@ mlx4_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr, { struct priv *priv = dev->data->dev_private; + if (mlx4_is_secondary()) + return; (void)vmdq; priv_lock(priv); 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)) goto end; priv_mac_addr_add(priv, index, (const uint8_t (*)[ETHER_ADDR_LEN]) @@ -3650,6 +4649,22 @@ end: priv_unlock(priv); } +/** + * 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); +} + /** * DPDK callback to enable promiscuous mode. * @@ -3663,6 +4678,8 @@ mlx4_promiscuous_enable(struct rte_eth_dev *dev) unsigned int i; int ret; + if (mlx4_is_secondary()) + return; priv_lock(priv); if (priv->promisc) { priv_unlock(priv); @@ -3709,6 +4726,8 @@ mlx4_promiscuous_disable(struct rte_eth_dev *dev) struct priv *priv = dev->data->dev_private; unsigned int i; + if (mlx4_is_secondary()) + return; priv_lock(priv); if (!priv->promisc) { priv_unlock(priv); @@ -3739,6 +4758,8 @@ mlx4_allmulticast_enable(struct rte_eth_dev *dev) unsigned int i; int ret; + if (mlx4_is_secondary()) + return; priv_lock(priv); if (priv->allmulti) { priv_unlock(priv); @@ -3785,6 +4806,8 @@ mlx4_allmulticast_disable(struct rte_eth_dev *dev) struct priv *priv = dev->data->dev_private; unsigned int i; + if (mlx4_is_secondary()) + return; priv_lock(priv); if (!priv->allmulti) { priv_unlock(priv); @@ -3813,30 +4836,42 @@ end: static int mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) { - struct priv *priv = dev->data->dev_private; - struct ibv_port_attr port_attr; - static const uint8_t width_mult[] = { - /* Multiplier values taken from devinfo.c in libibverbs. */ - 0, 1, 4, 0, 8, 0, 0, 0, 12, 0 + struct priv *priv = mlx4_get_priv(dev); + struct ethtool_cmd edata = { + .cmd = ETHTOOL_GSET }; + struct ifreq ifr; + struct rte_eth_link dev_link; + int link_speed = 0; + if (priv == NULL) + return -EINVAL; (void)wait_to_complete; - errno = ibv_query_port(priv->ctx, priv->port, &port_attr); - if (errno) { - WARN("port query failed: %s", strerror(errno)); + if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) { + WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno)); return -1; } - dev->data->dev_link = (struct rte_eth_link){ - .link_speed = (ibv_rate_to_mbps(mult_to_ibv_rate - (port_attr.active_speed)) * - width_mult[(port_attr.active_width % - sizeof(width_mult))]), - .link_duplex = ETH_LINK_FULL_DUPLEX, - .link_status = (port_attr.state == IBV_PORT_ACTIVE) - }; - if (memcmp(&port_attr, &priv->port_attr, sizeof(port_attr))) { + memset(&dev_link, 0, sizeof(dev_link)); + dev_link.link_status = ((ifr.ifr_flags & IFF_UP) && + (ifr.ifr_flags & IFF_RUNNING)); + ifr.ifr_data = (void *)&edata; + if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) { + WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s", + strerror(errno)); + return -1; + } + link_speed = ethtool_cmd_speed(&edata); + if (link_speed == -1) + dev_link.link_speed = 0; + else + 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. */ - priv->port_attr = port_attr; + dev->data->dev_link = dev_link; return 0; } /* Link status is still the same. */ @@ -3854,9 +4889,11 @@ mlx4_link_update_unlocked(struct rte_eth_dev *dev, int wait_to_complete) static int mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete) { - struct priv *priv = dev->data->dev_private; + struct priv *priv = mlx4_get_priv(dev); int ret; + if (priv == NULL) + return -EINVAL; priv_lock(priv); ret = mlx4_link_update_unlocked(dev, wait_to_complete); priv_unlock(priv); @@ -3889,6 +4926,8 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) uint16_t (*rx_func)(void *, struct rte_mbuf **, uint16_t) = mlx4_rx_burst; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; priv_lock(priv); /* Set kernel interface MTU first. */ if (priv_set_mtu(priv, mtu)) { @@ -3909,6 +4948,7 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) /* Reconfigure each RX queue. */ for (i = 0; (i != priv->rxqs_n); ++i) { struct rxq *rxq = (*priv->rxqs)[i]; + unsigned int mb_len; unsigned int max_frame_len; int sp; @@ -3918,7 +4958,9 @@ mlx4_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) * toggle scattered support (sp) if necessary. */ 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)); + mb_len = rte_pktmbuf_data_room_size(rxq->mp); + assert(mb_len >= RTE_PKTMBUF_HEADROOM); + sp = (max_frame_len > (mb_len - RTE_PKTMBUF_HEADROOM)); /* Provide new values to rxq_setup(). */ dev->data->dev_conf.rxmode.jumbo_frame = sp; dev->data->dev_conf.rxmode.max_rx_pkt_len = max_frame_len; @@ -3972,7 +5014,9 @@ mlx4_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) }; int ret; - ifr.ifr_data = ðpause; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; + ifr.ifr_data = (void *)ðpause; priv_lock(priv); if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) { ret = errno; @@ -4020,7 +5064,9 @@ mlx4_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) }; int ret; - ifr.ifr_data = ðpause; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; + 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)) @@ -4159,6 +5205,8 @@ mlx4_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) struct priv *priv = dev->data->dev_private; int ret; + if (mlx4_is_secondary()) + return -E_RTE_SECONDARY; priv_lock(priv); ret = vlan_filter_set(dev, vlan_id, on); priv_unlock(priv); @@ -4170,6 +5218,8 @@ 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, @@ -4180,6 +5230,7 @@ static const struct eth_dev_ops mlx4_dev_ops = { .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, @@ -4195,14 +5246,8 @@ static const struct eth_dev_ops mlx4_dev_ops = { .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, - .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 }; /** @@ -4256,22 +5301,25 @@ mlx4_ibv_device_to_pci_addr(const struct ibv_device *device, } /** - * Derive MAC address from port GID. + * Get MAC address by querying netdevice. * + * @param[in] priv + * struct priv for the requested device. * @param[out] mac * MAC address output buffer. - * @param port - * Physical port number. - * @param[in] gid - * Port GID. + * + * @return + * 0 on success, -1 on failure and errno is set. */ -static void -mac_from_gid(uint8_t (*mac)[ETHER_ADDR_LEN], uint32_t port, uint8_t *gid) +static int +priv_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN]) { - memcpy(&(*mac)[0], gid + 8, 3); - memcpy(&(*mac)[3], gid + 13, 3); - if (port == 1) - (*mac)[0] ^= 2; + struct ifreq request; + + if (priv_ifreq(priv, SIOCGIFHWADDR, &request)) + return -1; + memcpy(mac, request.ifr_hwaddr.sa_data, ETHER_ADDR_LEN); + return 0; } /* Support up to 32 adapters. */ @@ -4327,6 +5375,158 @@ mlx4_getenv_int(const char *name) return atoi(val); } +static void +mlx4_dev_link_status_handler(void *); +static void +mlx4_dev_interrupt_handler(struct rte_intr_handle *, void *); + +/** + * Link status handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + * + * @return + * Nonzero if the callback process can be called immediately. + */ +static int +priv_dev_link_status_handler(struct priv *priv, struct rte_eth_dev *dev) +{ + struct ibv_async_event event; + int port_change = 0; + int ret = 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) + port_change = 1; + else + DEBUG("event type %d on port %d not handled", + event.event_type, event.element.port_num); + ibv_ack_async_event(&event); + } + + if (port_change ^ priv->pending_alarm) { + struct rte_eth_link *link = &dev->data->dev_link; + + priv->pending_alarm = 0; + mlx4_link_update_unlocked(dev, 0); + if (((link->link_speed == 0) && link->link_status) || + ((link->link_speed != 0) && !link->link_status)) { + /* Inconsistent status, check again later. */ + priv->pending_alarm = 1; + rte_eal_alarm_set(MLX4_ALARM_TIMEOUT_US, + mlx4_dev_link_status_handler, + dev); + } else + ret = 1; + } + return ret; +} + +/** + * Handle delayed link status event. + * + * @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; + int ret; + + priv_lock(priv); + assert(priv->pending_alarm == 1); + ret = priv_dev_link_status_handler(priv, dev); + priv_unlock(priv); + if (ret) + _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC); +} + +/** + * Handle interrupts from the NIC. + * + * @param[in] intr_handle + * Interrupt handler. + * @param cb_arg + * Callback argument. + */ +static void +mlx4_dev_interrupt_handler(struct rte_intr_handle *intr_handle, void *cb_arg) +{ + struct rte_eth_dev *dev = cb_arg; + struct priv *priv = dev->data->dev_private; + int ret; + + (void)intr_handle; + priv_lock(priv); + ret = priv_dev_link_status_handler(priv, dev); + priv_unlock(priv); + if (ret) + _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC); +} + +/** + * Uninstall interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + */ +static void +priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev) +{ + if (!dev->data->dev_conf.intr_conf.lsc) + return; + rte_intr_callback_unregister(&priv->intr_handle, + mlx4_dev_interrupt_handler, + dev); + if (priv->pending_alarm) + rte_eal_alarm_cancel(mlx4_dev_link_status_handler, dev); + priv->pending_alarm = 0; + priv->intr_handle.fd = 0; + priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; +} + +/** + * Install interrupt handler. + * + * @param priv + * Pointer to private structure. + * @param dev + * Pointer to the rte_eth_dev structure. + */ +static void +priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev) +{ + int rc, flags; + + if (!dev->data->dev_conf.intr_conf.lsc) + return; + 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; + } else { + priv->intr_handle.fd = priv->ctx->async_fd; + priv->intr_handle.type = RTE_INTR_HANDLE_EXT; + rte_intr_callback_register(&priv->intr_handle, + mlx4_dev_interrupt_handler, + dev); + } +} + static struct eth_driver mlx4_driver; /** @@ -4344,7 +5544,7 @@ 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; @@ -4428,18 +5628,18 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) struct ibv_port_attr port_attr; struct ibv_pd *pd = NULL; struct priv *priv = NULL; - struct rte_eth_dev *eth_dev; -#if defined(INLINE_RECV) || defined(RSS_SUPPORT) + struct rte_eth_dev *eth_dev = NULL; +#ifdef HAVE_EXP_QUERY_DEVICE struct ibv_exp_device_attr exp_device_attr; -#endif +#endif /* HAVE_EXP_QUERY_DEVICE */ struct ether_addr mac; - union ibv_gid temp_gid; +#ifdef HAVE_EXP_QUERY_DEVICE + exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS; #ifdef RSS_SUPPORT - exp_device_attr.comp_mask = - (IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS | - IBV_EXP_DEVICE_ATTR_RSS_TBL_SZ); + 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); @@ -4453,10 +5653,17 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) ERROR("port query failed: %s", strerror(err)); goto port_error; } + + if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) { + ERROR("port %d is not configured in Ethernet mode", + port); + 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); @@ -4480,15 +5687,15 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) priv->ctx = ctx; priv->device_attr = device_attr; - priv->port_attr = port_attr; priv->port = port; priv->pd = pd; priv->mtu = ETHER_MTU; -#ifdef RSS_SUPPORT +#ifdef HAVE_EXP_QUERY_DEVICE if (ibv_exp_query_device(ctx, &exp_device_attr)) { - INFO("experimental ibv_exp_query_device"); + ERROR("ibv_exp_query_device() failed"); 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 & @@ -4515,6 +5722,19 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) exp_device_attr.max_rss_tbl_sz); #endif /* RSS_SUPPORT */ + priv->hw_csum = + ((exp_device_attr.exp_device_cap_flags & + IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) && + (exp_device_attr.exp_device_cap_flags & + IBV_EXP_DEVICE_RX_CSUM_IP_PKT)); + DEBUG("checksum offloading is %ssupported", + (priv->hw_csum ? "" : "not ")); + + priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags & + IBV_EXP_DEVICE_VXLAN_SUPPORT); + 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"); @@ -4540,15 +5760,16 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) priv->inl_recv_size); } #endif /* INLINE_RECV */ +#endif /* HAVE_EXP_QUERY_DEVICE */ (void)mlx4_getenv_int; priv->vf = vf; - if (ibv_query_gid(ctx, port, 0, &temp_gid)) { - ERROR("ibv_query_gid() failure"); + /* 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)); goto port_error; } - /* Configure the first MAC address by default. */ - mac_from_gid(&mac.addr_bytes, port, temp_gid.raw); INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x", priv->port, mac.addr_bytes[0], mac.addr_bytes[1], @@ -4558,7 +5779,7 @@ 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 @@ -4582,7 +5803,7 @@ 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"); @@ -4590,15 +5811,47 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) goto port_error; } - eth_dev->data->dev_private = priv; + /* Secondary processes have to use local storage for their + * private data as well as a copy of eth_dev->data, but this + * pointer must not be modified before burst functions are + * actually called. */ + if (mlx4_is_secondary()) { + struct mlx4_secondary_data *sd = + &mlx4_secondary_data[eth_dev->data->port_id]; + + sd->primary_priv = eth_dev->data->dev_private; + if (sd->primary_priv == NULL) { + ERROR("no private data for port %u", + eth_dev->data->port_id); + err = EINVAL; + goto port_error; + } + sd->shared_dev_data = eth_dev->data; + rte_spinlock_init(&sd->lock); + memcpy(sd->data.name, sd->shared_dev_data->name, + sizeof(sd->data.name)); + sd->data.dev_private = priv; + sd->data.rx_mbuf_alloc_failed = 0; + sd->data.mtu = ETHER_MTU; + sd->data.port_id = sd->shared_dev_data->port_id; + sd->data.mac_addrs = priv->mac; + eth_dev->tx_pkt_burst = mlx4_tx_burst_secondary_setup; + eth_dev->rx_pkt_burst = mlx4_rx_burst_secondary_setup; + } else { + eth_dev->data->dev_private = priv; + eth_dev->data->rx_mbuf_alloc_failed = 0; + eth_dev->data->mtu = ETHER_MTU; + eth_dev->data->mac_addrs = priv->mac; + } eth_dev->pci_dev = pci_dev; + + rte_eth_copy_pci_info(eth_dev, pci_dev); + eth_dev->driver = &mlx4_driver; - eth_dev->data->rx_mbuf_alloc_failed = 0; - eth_dev->data->mtu = ETHER_MTU; priv->dev = eth_dev; eth_dev->dev_ops = &mlx4_dev_ops; - eth_dev->data->mac_addrs = priv->mac; + TAILQ_INIT(ð_dev->link_intr_cbs); /* Bring Ethernet device up. */ DEBUG("forcing Ethernet interface up"); @@ -4611,6 +5864,8 @@ 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; } @@ -4638,22 +5893,16 @@ 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 @@ -4664,7 +5913,8 @@ static struct eth_driver mlx4_driver = { .pci_drv = { .name = MLX4_DRIVER_NAME, .id_table = mlx4_pci_id_map, - .devinit = mlx4_pci_devinit, + .probe = mlx4_pci_probe, + .drv_flags = RTE_PCI_DRV_INTR_LSC, }, .dev_private_size = sizeof(struct priv) }; @@ -4672,19 +5922,21 @@ static struct eth_driver mlx4_driver = { /** * 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 + * applications to use fork() safely for purposes other than + * using this PMD, which is not supported in forked processes. + */ + setenv("RDMAV_HUGEPAGES_SAFE", "1", 1); + ibv_fork_init(); rte_eal_pci_register(&mlx4_driver.pci_drv); - return 0; } -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) +DRIVER_EXPORT_NAME(net_mlx4, __COUNTER__); +DRIVER_REGISTER_PCI_TABLE(net_mlx4, mlx4_pci_id_map);