help On Fri, Oct 30, 2015 at 12:00 PM, <dev-request at dpdk.org> wrote:
> Send dev mailing list submissions to > dev at dpdk.org > > To subscribe or unsubscribe via the World Wide Web, visit > http://dpdk.org/ml/listinfo/dev > or, via email, send a message with subject or body 'help' to > dev-request at dpdk.org > > You can reach the person managing the list at > dev-owner at dpdk.org > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of dev digest..." > > > Today's Topics: > > 1. [PATCH v2 2/3] ethdev: change RETA type in > rte_eth_rss_reta_entry64 (Adrien Mazarguil) > 2. [PATCH v2 3/3] mlx5: RETA query/update support (Adrien Mazarguil) > 3. [PATCH 1/3] mlx4: improve RX performance with better > prefetching (Adrien Mazarguil) > 4. [PATCH 2/3] mlx4: do not complain about port state > (Adrien Mazarguil) > 5. [PATCH 3/3] mlx4: fix missing unlock in case of error > (Adrien Mazarguil) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Fri, 30 Oct 2015 19:58:56 +0100 > From: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > To: dev at dpdk.org > Subject: [dpdk-dev] [PATCH v2 2/3] ethdev: change RETA type in > rte_eth_rss_reta_entry64 > Message-ID: > <1446231537-8380-3-git-send-email-adrien.mazarguil at 6wind.com> > > From: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com> > > Several NICs can handle 512 entries in their RETA table, an 8 bit field is > not large enough for them. > > Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com> > --- > lib/librte_ether/rte_ethdev.h | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h > index 8a8c82b..560f4e8 100644 > --- a/lib/librte_ether/rte_ethdev.h > +++ b/lib/librte_ether/rte_ethdev.h > @@ -516,7 +516,7 @@ struct rte_eth_mirror_conf { > struct rte_eth_rss_reta_entry64 { > uint64_t mask; > /**< Mask bits indicate which entries need to be updated/queried. > */ > - uint8_t reta[RTE_RETA_GROUP_SIZE]; > + uint16_t reta[RTE_RETA_GROUP_SIZE]; > /**< Group of 64 redirection table entries. */ > }; > > -- > 2.1.0 > > > > ------------------------------ > > Message: 2 > Date: Fri, 30 Oct 2015 19:58:57 +0100 > From: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > To: dev at dpdk.org > Subject: [dpdk-dev] [PATCH v2 3/3] mlx5: RETA query/update support > Message-ID: > <1446231537-8380-4-git-send-email-adrien.mazarguil at 6wind.com> > > From: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com> > > ConnectX-4 is able to use indirection tables size of power of two, but > with the current API it is impossible to predict its size, so to simplify, > for any query/update RETA command, the indirection table is modified to use > the maximum value. > > A port stop/start must be done to apply the new RETA configuration. > > Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com> > --- > drivers/net/mlx5/mlx5.c | 4 + > drivers/net/mlx5/mlx5.h | 7 ++ > drivers/net/mlx5/mlx5_ethdev.c | 29 ++++++++ > drivers/net/mlx5/mlx5_rss.c | 163 > +++++++++++++++++++++++++++++++++++++++++ > drivers/net/mlx5/mlx5_rxq.c | 53 ++------------ > drivers/net/mlx5/mlx5_utils.h | 20 +++++ > 6 files changed, 231 insertions(+), 45 deletions(-) > > diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c > index 9636588..43a40d7 100644 > --- a/drivers/net/mlx5/mlx5.c > +++ b/drivers/net/mlx5/mlx5.c > @@ -133,6 +133,8 @@ mlx5_dev_close(struct rte_eth_dev *dev) > rte_free((*priv->rss_conf)[i]); > rte_free(priv->rss_conf); > } > + if (priv->reta_idx != NULL) > + rte_free(priv->reta_idx); > priv_unlock(priv); > memset(priv, 0, sizeof(*priv)); > } > @@ -160,6 +162,8 @@ static const struct eth_dev_ops mlx5_dev_ops = { > .mac_addr_remove = mlx5_mac_addr_remove, > .mac_addr_add = mlx5_mac_addr_add, > .mtu_set = mlx5_dev_set_mtu, > + .reta_update = mlx5_dev_rss_reta_update, > + .reta_query = mlx5_dev_rss_reta_query, > .rss_hash_update = mlx5_rss_hash_update, > .rss_hash_conf_get = mlx5_rss_hash_conf_get, > }; > diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h > index 0daacc8..b84d31d 100644 > --- a/drivers/net/mlx5/mlx5.h > +++ b/drivers/net/mlx5/mlx5.h > @@ -118,6 +118,8 @@ struct priv { > /* RSS configuration array indexed by hash RX queue type. */ > struct rte_eth_rss_conf *(*rss_conf)[]; > struct rte_intr_handle intr_handle; /* Interrupt handler. */ > + unsigned int (*reta_idx)[]; /* RETA index table. */ > + unsigned int reta_idx_n; /* RETA index size. */ > rte_spinlock_t lock; /* Lock for control functions. */ > }; > > @@ -184,6 +186,11 @@ int rss_hash_rss_conf_new_key(struct priv *, const > uint8_t *, unsigned int, > uint64_t); > int mlx5_rss_hash_update(struct rte_eth_dev *, struct rte_eth_rss_conf *); > int mlx5_rss_hash_conf_get(struct rte_eth_dev *, struct rte_eth_rss_conf > *); > +int priv_rss_reta_index_resize(struct priv *, unsigned int); > +int mlx5_dev_rss_reta_query(struct rte_eth_dev *, > + struct rte_eth_rss_reta_entry64 *, uint16_t); > +int mlx5_dev_rss_reta_update(struct rte_eth_dev *, > + struct rte_eth_rss_reta_entry64 *, uint16_t); > > /* mlx5_rxmode.c */ > > diff --git a/drivers/net/mlx5/mlx5_ethdev.c > b/drivers/net/mlx5/mlx5_ethdev.c > index 84e877c..1159fa3 100644 > --- a/drivers/net/mlx5/mlx5_ethdev.c > +++ b/drivers/net/mlx5/mlx5_ethdev.c > @@ -410,6 +410,9 @@ dev_configure(struct rte_eth_dev *dev) > struct priv *priv = dev->data->dev_private; > unsigned int rxqs_n = dev->data->nb_rx_queues; > unsigned int txqs_n = dev->data->nb_tx_queues; > + unsigned int i; > + unsigned int j; > + unsigned int reta_idx_n; > > priv->rxqs = (void *)dev->data->rx_queues; > priv->txqs = (void *)dev->data->tx_queues; > @@ -418,11 +421,31 @@ dev_configure(struct rte_eth_dev *dev) > (void *)dev, priv->txqs_n, txqs_n); > priv->txqs_n = txqs_n; > } > + if (rxqs_n > priv->ind_table_max_size) { > + ERROR("cannot handle this many RX queues (%u)", rxqs_n); > + return EINVAL; > + } > if (rxqs_n == priv->rxqs_n) > return 0; > INFO("%p: RX queues number update: %u -> %u", > (void *)dev, priv->rxqs_n, rxqs_n); > priv->rxqs_n = rxqs_n; > + /* If the requested number of RX queues is not a power of two, use > the > + * maximum indirection table size for better balancing. > + * The result is always rounded to the next power of two. */ > + reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ? > + priv->ind_table_max_size : > + rxqs_n)); > + if (priv_rss_reta_index_resize(priv, reta_idx_n)) > + return ENOMEM; > + /* When the number of RX queues is not a power of two, the > remaining > + * table entries are padded with reused WQs and hashes are not > spread > + * uniformly. */ > + for (i = 0, j = 0; (i != reta_idx_n); ++i) { > + (*priv->reta_idx)[i] = j; > + if (++j == rxqs_n) > + j = 0; > + } > return 0; > } > > @@ -494,6 +517,12 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct > rte_eth_dev_info *info) > 0); > if (priv_get_ifname(priv, &ifname) == 0) > info->if_index = if_nametoindex(ifname); > + /* FIXME: RETA update/query API expects the callee to know the > size of > + * the indirection table, for this PMD the size varies depending on > + * the number of RX queues, it becomes impossible to find the > correct > + * size if it is not fixed. > + * The API should be updated to solve this problem. */ > + info->reta_size = priv->ind_table_max_size; > priv_unlock(priv); > } > > diff --git a/drivers/net/mlx5/mlx5_rss.c b/drivers/net/mlx5/mlx5_rss.c > index bf19aca..7eb688a 100644 > --- a/drivers/net/mlx5/mlx5_rss.c > +++ b/drivers/net/mlx5/mlx5_rss.c > @@ -211,3 +211,166 @@ mlx5_rss_hash_conf_get(struct rte_eth_dev *dev, > priv_unlock(priv); > return 0; > } > + > +/** > + * Allocate/reallocate RETA index table. > + * > + * @param priv > + * Pointer to private structure. > + * @praram reta_size > + * The size of the array to allocate. > + * > + * @return > + * 0 on success, errno value on failure. > + */ > +int > +priv_rss_reta_index_resize(struct priv *priv, unsigned int reta_size) > +{ > + void *mem; > + unsigned int old_size = priv->reta_idx_n; > + > + if (priv->reta_idx_n == reta_size) > + return 0; > + > + mem = rte_realloc(priv->reta_idx, > + reta_size * sizeof((*priv->reta_idx)[0]), 0); > + if (!mem) > + return ENOMEM; > + priv->reta_idx = mem; > + priv->reta_idx_n = reta_size; > + > + if (old_size < reta_size) > + memset(&(*priv->reta_idx)[old_size], 0, > + (reta_size - old_size) * > + sizeof((*priv->reta_idx)[0])); > + return 0; > +} > + > +/** > + * Query RETA table. > + * > + * @param priv > + * Pointer to private structure. > + * @param[in, out] reta_conf > + * Pointer to the first RETA configuration structure. > + * @param reta_size > + * Number of entries. > + * > + * @return > + * 0 on success, errno value on failure. > + */ > +static int > +priv_dev_rss_reta_query(struct priv *priv, > + struct rte_eth_rss_reta_entry64 *reta_conf, > + unsigned int reta_size) > +{ > + unsigned int idx; > + unsigned int i; > + int ret; > + > + /* See RETA comment in mlx5_dev_infos_get(). */ > + ret = priv_rss_reta_index_resize(priv, priv->ind_table_max_size); > + if (ret) > + return ret; > + > + /* Fill each entry of the table even if its bit is not set. */ > + for (idx = 0, i = 0; (i != reta_size); ++i) { > + idx = i / RTE_RETA_GROUP_SIZE; > + reta_conf[idx].reta[i % RTE_RETA_GROUP_SIZE] = > + (*priv->reta_idx)[i]; > + } > + return 0; > +} > + > +/** > + * Update RETA table. > + * > + * @param priv > + * Pointer to private structure. > + * @param[in] reta_conf > + * Pointer to the first RETA configuration structure. > + * @param reta_size > + * Number of entries. > + * > + * @return > + * 0 on success, errno value on failure. > + */ > +static int > +priv_dev_rss_reta_update(struct priv *priv, > + struct rte_eth_rss_reta_entry64 *reta_conf, > + unsigned int reta_size) > +{ > + unsigned int idx; > + unsigned int i; > + unsigned int pos; > + int ret; > + > + /* See RETA comment in mlx5_dev_infos_get(). */ > + ret = priv_rss_reta_index_resize(priv, priv->ind_table_max_size); > + if (ret) > + return ret; > + > + for (idx = 0, i = 0; (i != reta_size); ++i) { > + idx = i / RTE_RETA_GROUP_SIZE; > + pos = i % RTE_RETA_GROUP_SIZE; > + if (((reta_conf[idx].mask >> i) & 0x1) == 0) > + continue; > + assert(reta_conf[idx].reta[pos] < priv->rxqs_n); > + (*priv->reta_idx)[i] = reta_conf[idx].reta[pos]; > + } > + return 0; > +} > + > +/** > + * DPDK callback to get the RETA indirection table. > + * > + * @param dev > + * Pointer to Ethernet device structure. > + * @param reta_conf > + * Pointer to RETA configuration structure array. > + * @param reta_size > + * Size of the RETA table. > + * > + * @return > + * 0 on success, negative errno value on failure. > + */ > +int > +mlx5_dev_rss_reta_query(struct rte_eth_dev *dev, > + struct rte_eth_rss_reta_entry64 *reta_conf, > + uint16_t reta_size) > +{ > + int ret; > + struct priv *priv = dev->data->dev_private; > + > + priv_lock(priv); > + ret = priv_dev_rss_reta_query(priv, reta_conf, reta_size); > + priv_unlock(priv); > + return -ret; > +} > + > +/** > + * DPDK callback to update the RETA indirection table. > + * > + * @param dev > + * Pointer to Ethernet device structure. > + * @param reta_conf > + * Pointer to RETA configuration structure array. > + * @param reta_size > + * Size of the RETA table. > + * > + * @return > + * 0 on success, negative errno value on failure. > + */ > +int > +mlx5_dev_rss_reta_update(struct rte_eth_dev *dev, > + struct rte_eth_rss_reta_entry64 *reta_conf, > + uint16_t reta_size) > +{ > + int ret; > + struct priv *priv = dev->data->dev_private; > + > + priv_lock(priv); > + ret = priv_dev_rss_reta_update(priv, reta_conf, reta_size); > + priv_unlock(priv); > + return -ret; > +} > diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c > index 084bf41..3d7ae7e 100644 > --- a/drivers/net/mlx5/mlx5_rxq.c > +++ b/drivers/net/mlx5/mlx5_rxq.c > @@ -259,26 +259,6 @@ hash_rxq_flow_attr(const struct hash_rxq *hash_rxq, > } > > /** > - * Return nearest power of two above input value. > - * > - * @param v > - * Input value. > - * > - * @return > - * Nearest power of two above input value. > - */ > -static unsigned int > -log2above(unsigned int v) > -{ > - unsigned int l; > - unsigned int r; > - > - for (l = 0, r = 0; (v >> 1); ++l, v >>= 1) > - r |= (v & 1); > - return (l + r); > -} > - > -/** > * Return the type corresponding to the n'th bit set. > * > * @param table > @@ -360,14 +340,7 @@ priv_make_ind_table_init(struct priv *priv, > int > priv_create_hash_rxqs(struct priv *priv) > { > - /* If the requested number of WQs is not a power of two, use the > - * maximum indirection table size for better balancing. > - * The result is always rounded to the next power of two. */ > - unsigned int wqs_n = > - (1 << log2above((priv->rxqs_n & (priv->rxqs_n - 1)) ? > - priv->ind_table_max_size : > - priv->rxqs_n)); > - struct ibv_exp_wq *wqs[wqs_n]; > + struct ibv_exp_wq *wqs[priv->reta_idx_n]; > struct ind_table_init ind_table_init[IND_TABLE_INIT_N]; > unsigned int ind_tables_n = > priv_make_ind_table_init(priv, &ind_table_init); > @@ -393,25 +366,15 @@ priv_create_hash_rxqs(struct priv *priv) > " indirection table cannot be created"); > return EINVAL; > } > - if ((wqs_n < priv->rxqs_n) || (wqs_n > priv->ind_table_max_size)) { > - ERROR("cannot handle this many RX queues (%u)", > priv->rxqs_n); > - err = ERANGE; > - goto error; > - } > - if (wqs_n != priv->rxqs_n) { > + if (priv->rxqs_n & (priv->rxqs_n - 1)) { > INFO("%u RX queues are configured, consider rounding this" > " number to the next power of two for better > balancing", > priv->rxqs_n); > - DEBUG("indirection table extended to assume %u WQs", > wqs_n); > - } > - /* When the number of RX queues is not a power of two, the > remaining > - * table entries are padded with reused WQs and hashes are not > spread > - * uniformly. */ > - for (i = 0, j = 0; (i != wqs_n); ++i) { > - wqs[i] = (*priv->rxqs)[j]->wq; > - if (++j == priv->rxqs_n) > - j = 0; > + DEBUG("indirection table extended to assume %u WQs", > + priv->reta_idx_n); > } > + for (i = 0; (i != priv->reta_idx_n); ++i) > + wqs[i] = (*priv->rxqs)[(*priv->reta_idx)[i]]->wq; > /* Get number of hash RX queues to configure. */ > for (i = 0, hash_rxqs_n = 0; (i != ind_tables_n); ++i) > hash_rxqs_n += ind_table_init[i].hash_types_n; > @@ -436,8 +399,8 @@ priv_create_hash_rxqs(struct priv *priv) > unsigned int ind_tbl_size = ind_table_init[i].max_size; > struct ibv_exp_rwq_ind_table *ind_table; > > - if (wqs_n < ind_tbl_size) > - ind_tbl_size = wqs_n; > + if (priv->reta_idx_n < ind_tbl_size) > + ind_tbl_size = priv->reta_idx_n; > ind_init_attr.log_ind_tbl_size = log2above(ind_tbl_size); > errno = 0; > ind_table = ibv_exp_create_rwq_ind_table(priv->ctx, > diff --git a/drivers/net/mlx5/mlx5_utils.h b/drivers/net/mlx5/mlx5_utils.h > index f1fad18..9b5e86a 100644 > --- a/drivers/net/mlx5/mlx5_utils.h > +++ b/drivers/net/mlx5/mlx5_utils.h > @@ -161,4 +161,24 @@ pmd_drv_log_basename(const char *s) > \ > snprintf(name, sizeof(name), __VA_ARGS__) > > +/** > + * Return nearest power of two above input value. > + * > + * @param v > + * Input value. > + * > + * @return > + * Nearest power of two above input value. > + */ > +static inline unsigned int > +log2above(unsigned int v) > +{ > + unsigned int l; > + unsigned int r; > + > + for (l = 0, r = 0; (v >> 1); ++l, v >>= 1) > + r |= (v & 1); > + return (l + r); > +} > + > #endif /* RTE_PMD_MLX5_UTILS_H_ */ > -- > 2.1.0 > > > > ------------------------------ > > Message: 3 > Date: Fri, 30 Oct 2015 20:00:07 +0100 > From: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > To: dev at dpdk.org > Subject: [dpdk-dev] [PATCH 1/3] mlx4: improve RX performance with > better prefetching > Message-ID: > <1446231609-8463-1-git-send-email-adrien.mazarguil at 6wind.com> > > From: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com> > > Prefetching initial bytes of mbuf structures earlier and in two cache lines > instead of one improves performance of mlx4_rx_burst(), which accesses the > mbuf->next field not present in the first 128 bytes. > > Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com> > --- > drivers/net/mlx4/mlx4.c | 11 ++++++----- > 1 file changed, 6 insertions(+), 5 deletions(-) > > diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c > index cb08ee8..4cf2f7d 100644 > --- a/drivers/net/mlx4/mlx4.c > +++ b/drivers/net/mlx4/mlx4.c > @@ -2824,6 +2824,12 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf > **pkts, uint16_t pkts_n) > assert(wr->num_sge == 1); > assert(elts_head < rxq->elts_n); > assert(rxq->elts_head < rxq->elts_n); > + /* > + * Fetch initial bytes of packet descriptor into a > + * cacheline while allocating rep. > + */ > + rte_prefetch0(seg); > + rte_prefetch0(&seg->cacheline1); > ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL, > &flags); > if (unlikely(ret < 0)) { > @@ -2861,11 +2867,6 @@ mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf > **pkts, uint16_t pkts_n) > if (ret == 0) > break; > len = ret; > - /* > - * Fetch initial bytes of packet descriptor into a > - * cacheline while allocating rep. > - */ > - rte_prefetch0(seg); > rep = __rte_mbuf_raw_alloc(rxq->mp); > if (unlikely(rep == NULL)) { > /* > -- > 2.1.0 > > > > ------------------------------ > > Message: 4 > Date: Fri, 30 Oct 2015 20:00:08 +0100 > From: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > To: dev at dpdk.org > Subject: [dpdk-dev] [PATCH 2/3] mlx4: do not complain about port state > Message-ID: > <1446231609-8463-2-git-send-email-adrien.mazarguil at 6wind.com> > > Port is considering inactive when the related netdevice is down. There is > no > reason to warn about it (and confuse users) since it's automatically > brought > up later by the PMD. > > Signed-off-by: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > --- > drivers/net/mlx4/mlx4.c | 6 +++--- > 1 file changed, 3 insertions(+), 3 deletions(-) > > diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c > index 4cf2f7d..3b25492 100644 > --- a/drivers/net/mlx4/mlx4.c > +++ b/drivers/net/mlx4/mlx4.c > @@ -4985,9 +4985,9 @@ mlx4_pci_devinit(struct rte_pci_driver *pci_drv, > struct rte_pci_device *pci_dev) > 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); > -- > 2.1.0 > > > > ------------------------------ > > Message: 5 > Date: Fri, 30 Oct 2015 20:00:09 +0100 > From: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > To: dev at dpdk.org > Subject: [dpdk-dev] [PATCH 3/3] mlx4: fix missing unlock in case of > error > Message-ID: > <1446231609-8463-3-git-send-email-adrien.mazarguil at 6wind.com> > > Fixes: 7fae69eeff13 ("mlx4: new poll mode driver") > > Signed-off-by: Adrien Mazarguil <adrien.mazarguil at 6wind.com> > --- > drivers/net/mlx4/mlx4.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c > index 3b25492..691e0e5 100644 > --- a/drivers/net/mlx4/mlx4.c > +++ b/drivers/net/mlx4/mlx4.c > @@ -3652,6 +3652,7 @@ 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); > -- > 2.1.0 > > > > End of dev Digest, Vol 64, Issue 177 > ************************************ >
