net/bonding: provide default Rx/Tx configuration
authorIvan Malov <ivan.malov@oktetlabs.ru>
Wed, 5 Sep 2018 09:13:37 +0000 (10:13 +0100)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 27 Sep 2018 23:41:03 +0000 (01:41 +0200)
Default Rx/Tx configuration has become a helpful
resource for applications relying on the optimal
values to make rte_eth_rxconf and rte_eth_txconf
structures. These structures can then be tweaked.

Default configuration is also used internally by
rte_eth_rx_queue_setup or rte_eth_tx_queue_setup
API calls when NULL pointer is passed by callers
with the argument for custom queue configuration.

The use cases of bonding driver may also benefit
from exercising default settings in the same way.

Restructure the code to collect various settings
from slave ports and make it possible to combine
default Rx/Tx configuration of these devices and
report it to the callers of rte_eth_dev_info_get.

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Acked-by: Chas Williams <chas3@att.com>
drivers/net/bonding/rte_eth_bond_api.c
drivers/net/bonding/rte_eth_bond_pmd.c
drivers/net/bonding/rte_eth_bond_private.h

index 8bc04cf..206a5c7 100644 (file)
@@ -269,6 +269,136 @@ slave_rte_flow_prepare(uint16_t slave_id, struct bond_dev_private *internals)
        return 0;
 }
 
+static void
+eth_bond_slave_inherit_dev_info_rx_first(struct bond_dev_private *internals,
+                                        const struct rte_eth_dev_info *di)
+{
+       struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
+
+       internals->reta_size = di->reta_size;
+
+       /* Inherit Rx offload capabilities from the first slave device */
+       internals->rx_offload_capa = di->rx_offload_capa;
+       internals->rx_queue_offload_capa = di->rx_queue_offload_capa;
+       internals->flow_type_rss_offloads = di->flow_type_rss_offloads;
+
+       /* Inherit maximum Rx packet size from the first slave device */
+       internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
+
+       /* Inherit default Rx queue settings from the first slave device */
+       memcpy(rxconf_i, &di->default_rxconf, sizeof(*rxconf_i));
+
+       /*
+        * Turn off descriptor prefetch and writeback by default for all
+        * slave devices. Applications may tweak this setting if need be.
+        */
+       rxconf_i->rx_thresh.pthresh = 0;
+       rxconf_i->rx_thresh.hthresh = 0;
+       rxconf_i->rx_thresh.wthresh = 0;
+
+       /* Setting this to zero should effectively enable default values */
+       rxconf_i->rx_free_thresh = 0;
+
+       /* Disable deferred start by default for all slave devices */
+       rxconf_i->rx_deferred_start = 0;
+}
+
+static void
+eth_bond_slave_inherit_dev_info_tx_first(struct bond_dev_private *internals,
+                                        const struct rte_eth_dev_info *di)
+{
+       struct rte_eth_txconf *txconf_i = &internals->default_txconf;
+
+       /* Inherit Tx offload capabilities from the first slave device */
+       internals->tx_offload_capa = di->tx_offload_capa;
+       internals->tx_queue_offload_capa = di->tx_queue_offload_capa;
+
+       /* Inherit default Tx queue settings from the first slave device */
+       memcpy(txconf_i, &di->default_txconf, sizeof(*txconf_i));
+
+       /*
+        * Turn off descriptor prefetch and writeback by default for all
+        * slave devices. Applications may tweak this setting if need be.
+        */
+       txconf_i->tx_thresh.pthresh = 0;
+       txconf_i->tx_thresh.hthresh = 0;
+       txconf_i->tx_thresh.wthresh = 0;
+
+       /*
+        * Setting these parameters to zero assumes that default
+        * values will be configured implicitly by slave devices.
+        */
+       txconf_i->tx_free_thresh = 0;
+       txconf_i->tx_rs_thresh = 0;
+
+       /* Disable deferred start by default for all slave devices */
+       txconf_i->tx_deferred_start = 0;
+}
+
+static void
+eth_bond_slave_inherit_dev_info_rx_next(struct bond_dev_private *internals,
+                                       const struct rte_eth_dev_info *di)
+{
+       struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
+       const struct rte_eth_rxconf *rxconf = &di->default_rxconf;
+
+       internals->rx_offload_capa &= di->rx_offload_capa;
+       internals->rx_queue_offload_capa &= di->rx_queue_offload_capa;
+       internals->flow_type_rss_offloads &= di->flow_type_rss_offloads;
+
+       /*
+        * If at least one slave device suggests enabling this
+        * setting by default, enable it for all slave devices
+        * since disabling it may not be necessarily supported.
+        */
+       if (rxconf->rx_drop_en == 1)
+               rxconf_i->rx_drop_en = 1;
+
+       /*
+        * Adding a new slave device may cause some of previously inherited
+        * offloads to be withdrawn from the internal rx_queue_offload_capa
+        * value. Thus, the new internal value of default Rx queue offloads
+        * has to be masked by rx_queue_offload_capa to make sure that only
+        * commonly supported offloads are preserved from both the previous
+        * value and the value being inhereted from the new slave device.
+        */
+       rxconf_i->offloads = (rxconf_i->offloads | rxconf->offloads) &
+                            internals->rx_queue_offload_capa;
+
+       /*
+        * RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
+        * the power of 2, the lower one is GCD
+        */
+       if (internals->reta_size > di->reta_size)
+               internals->reta_size = di->reta_size;
+
+       if (!internals->max_rx_pktlen &&
+           di->max_rx_pktlen < internals->candidate_max_rx_pktlen)
+               internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
+}
+
+static void
+eth_bond_slave_inherit_dev_info_tx_next(struct bond_dev_private *internals,
+                                       const struct rte_eth_dev_info *di)
+{
+       struct rte_eth_txconf *txconf_i = &internals->default_txconf;
+       const struct rte_eth_txconf *txconf = &di->default_txconf;
+
+       internals->tx_offload_capa &= di->tx_offload_capa;
+       internals->tx_queue_offload_capa &= di->tx_queue_offload_capa;
+
+       /*
+        * Adding a new slave device may cause some of previously inherited
+        * offloads to be withdrawn from the internal tx_queue_offload_capa
+        * value. Thus, the new internal value of default Tx queue offloads
+        * has to be masked by tx_queue_offload_capa to make sure that only
+        * commonly supported offloads are preserved from both the previous
+        * value and the value being inhereted from the new slave device.
+        */
+       txconf_i->offloads = (txconf_i->offloads | txconf->offloads) &
+                            internals->tx_queue_offload_capa;
+}
+
 static int
 __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
 {
@@ -326,34 +456,11 @@ __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
                internals->nb_rx_queues = slave_eth_dev->data->nb_rx_queues;
                internals->nb_tx_queues = slave_eth_dev->data->nb_tx_queues;
 
-               internals->reta_size = dev_info.reta_size;
-
-               /* Take the first dev's offload capabilities */
-               internals->rx_offload_capa = dev_info.rx_offload_capa;
-               internals->tx_offload_capa = dev_info.tx_offload_capa;
-               internals->rx_queue_offload_capa = dev_info.rx_queue_offload_capa;
-               internals->tx_queue_offload_capa = dev_info.tx_queue_offload_capa;
-               internals->flow_type_rss_offloads = dev_info.flow_type_rss_offloads;
-
-               /* Inherit first slave's max rx packet size */
-               internals->candidate_max_rx_pktlen = dev_info.max_rx_pktlen;
-
+               eth_bond_slave_inherit_dev_info_rx_first(internals, &dev_info);
+               eth_bond_slave_inherit_dev_info_tx_first(internals, &dev_info);
        } else {
-               internals->rx_offload_capa &= dev_info.rx_offload_capa;
-               internals->tx_offload_capa &= dev_info.tx_offload_capa;
-               internals->rx_queue_offload_capa &= dev_info.rx_queue_offload_capa;
-               internals->tx_queue_offload_capa &= dev_info.tx_queue_offload_capa;
-               internals->flow_type_rss_offloads &= dev_info.flow_type_rss_offloads;
-
-               /* RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
-                * the power of 2, the lower one is GCD
-                */
-               if (internals->reta_size > dev_info.reta_size)
-                       internals->reta_size = dev_info.reta_size;
-
-               if (!internals->max_rx_pktlen &&
-                   dev_info.max_rx_pktlen < internals->candidate_max_rx_pktlen)
-                       internals->candidate_max_rx_pktlen = dev_info.max_rx_pktlen;
+               eth_bond_slave_inherit_dev_info_rx_next(internals, &dev_info);
+               eth_bond_slave_inherit_dev_info_tx_next(internals, &dev_info);
        }
 
        bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf &=
index 86e78bd..07eff60 100644 (file)
@@ -2248,6 +2248,11 @@ bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        dev_info->max_rx_queues = max_nb_rx_queues;
        dev_info->max_tx_queues = max_nb_tx_queues;
 
+       memcpy(&dev_info->default_rxconf, &internals->default_rxconf,
+              sizeof(dev_info->default_rxconf));
+       memcpy(&dev_info->default_txconf, &internals->default_txconf,
+              sizeof(dev_info->default_txconf));
+
        /**
         * If dedicated hw queues enabled for link bonding device in LACP mode
         * then we need to reduce the maximum number of data path queues by 1.
@@ -3068,6 +3073,11 @@ bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
        /* Initially allow to choose any offload type */
        internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
 
+       memset(&internals->default_rxconf, 0,
+              sizeof(internals->default_rxconf));
+       memset(&internals->default_txconf, 0,
+              sizeof(internals->default_txconf));
+
        memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
        memset(internals->slaves, 0, sizeof(internals->slaves));
 
index 43e0e44..d12a0eb 100644 (file)
@@ -160,6 +160,9 @@ struct bond_dev_private {
        /** Bit mask of RSS offloads, the bit offset also means flow type */
        uint64_t flow_type_rss_offloads;
 
+       struct rte_eth_rxconf default_rxconf;   /**< Default RxQ conf. */
+       struct rte_eth_txconf default_txconf;   /**< Default TxQ conf. */
+
        uint16_t reta_size;
        struct rte_eth_rss_reta_entry64 reta_conf[ETH_RSS_RETA_SIZE_512 /
                        RTE_RETA_GROUP_SIZE];