net/fm10k: convert to new Rx offloads API
authorWei Dai <wei.dai@intel.com>
Wed, 28 Mar 2018 08:00:36 +0000 (16:00 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 13 Apr 2018 22:40:21 +0000 (00:40 +0200)
Ethdev Rx offloads API has changed since:
commit ce17eddefc20 ("ethdev: introduce Rx queue offloads API")
This commit support the new Rx offloads API.

Signed-off-by: Wei Dai <wei.dai@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
drivers/net/fm10k/fm10k.h
drivers/net/fm10k/fm10k_ethdev.c
drivers/net/fm10k/fm10k_rxtx_vec.c

index 30dad3e..d3c80c2 100644 (file)
@@ -180,6 +180,7 @@ struct fm10k_rx_queue {
        uint8_t drop_en;
        uint8_t rx_deferred_start; /* don't start this queue in dev start. */
        uint16_t rx_ftag_en; /* indicates FTAG RX supported */
+       uint64_t offloads; /* offloads of DEV_RX_OFFLOAD_* */
 };
 
 /*
index 9423761..d2c47f6 100644 (file)
@@ -60,6 +60,10 @@ static void fm10k_set_tx_function(struct rte_eth_dev *dev);
 static int fm10k_check_ftag(struct rte_devargs *devargs);
 static int fm10k_link_update(struct rte_eth_dev *dev, int wait_to_complete);
 
+static void fm10k_dev_infos_get(struct rte_eth_dev *dev,
+                               struct rte_eth_dev_info *dev_info);
+static uint64_t fm10k_get_rx_queue_offloads_capa(struct rte_eth_dev *dev);
+static uint64_t fm10k_get_rx_port_offloads_capa(struct rte_eth_dev *dev);
 struct fm10k_xstats_name_off {
        char name[RTE_ETH_XSTATS_NAME_SIZE];
        unsigned offset;
@@ -441,11 +445,21 @@ static int
 fm10k_dev_configure(struct rte_eth_dev *dev)
 {
        int ret;
+       struct rte_eth_dev_info dev_info;
+       uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
 
        PMD_INIT_FUNC_TRACE();
 
-       if (dev->data->dev_conf.rxmode.hw_strip_crc == 0)
+       if ((rx_offloads & DEV_RX_OFFLOAD_CRC_STRIP) == 0)
                PMD_INIT_LOG(WARNING, "fm10k always strip CRC");
+
+       fm10k_dev_infos_get(dev, &dev_info);
+       if ((rx_offloads & dev_info.rx_offload_capa) != rx_offloads) {
+               PMD_DRV_LOG(ERR, "Some Rx offloads are not supported "
+                           "requested 0x%" PRIx64 " supported 0x%" PRIx64,
+                           rx_offloads, dev_info.rx_offload_capa);
+               return -ENOTSUP;
+       }
        /* multipe queue mode checking */
        ret  = fm10k_check_mq_mode(dev);
        if (ret != 0) {
@@ -454,6 +468,8 @@ fm10k_dev_configure(struct rte_eth_dev *dev)
                return ret;
        }
 
+       dev->data->scattered_rx = 0;
+
        return 0;
 }
 
@@ -756,7 +772,7 @@ fm10k_dev_rx_init(struct rte_eth_dev *dev)
                /* It adds dual VLAN length for supporting dual VLAN */
                if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
                                2 * FM10K_VLAN_TAG_SIZE) > buf_size ||
-                       dev->data->dev_conf.rxmode.enable_scatter) {
+                       rxq->offloads & DEV_RX_OFFLOAD_SCATTER) {
                        uint32_t reg;
                        dev->data->scattered_rx = 1;
                        reg = FM10K_READ_REG(hw, FM10K_SRRCTL(i));
@@ -1389,11 +1405,9 @@ fm10k_dev_infos_get(struct rte_eth_dev *dev,
        dev_info->vmdq_queue_base    = 0;
        dev_info->max_vmdq_pools     = ETH_32_POOLS;
        dev_info->vmdq_queue_num     = FM10K_MAX_QUEUES_PF;
-       dev_info->rx_offload_capa =
-               DEV_RX_OFFLOAD_VLAN_STRIP |
-               DEV_RX_OFFLOAD_IPV4_CKSUM |
-               DEV_RX_OFFLOAD_UDP_CKSUM  |
-               DEV_RX_OFFLOAD_TCP_CKSUM;
+       dev_info->rx_queue_offload_capa = fm10k_get_rx_queue_offloads_capa(dev);
+       dev_info->rx_offload_capa = fm10k_get_rx_port_offloads_capa(dev) |
+                                   dev_info->rx_queue_offload_capa;
        dev_info->tx_offload_capa =
                DEV_TX_OFFLOAD_VLAN_INSERT |
                DEV_TX_OFFLOAD_IPV4_CKSUM  |
@@ -1412,6 +1426,7 @@ fm10k_dev_infos_get(struct rte_eth_dev *dev,
                },
                .rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(0),
                .rx_drop_en = 0,
+               .offloads = 0,
        };
 
        dev_info->default_txconf = (struct rte_eth_txconf) {
@@ -1571,19 +1586,22 @@ static int
 fm10k_vlan_offload_set(struct rte_eth_dev *dev, int mask)
 {
        if (mask & ETH_VLAN_STRIP_MASK) {
-               if (!dev->data->dev_conf.rxmode.hw_vlan_strip)
+               if (!(dev->data->dev_conf.rxmode.offloads &
+                       DEV_RX_OFFLOAD_VLAN_STRIP))
                        PMD_INIT_LOG(ERR, "VLAN stripping is "
                                        "always on in fm10k");
        }
 
        if (mask & ETH_VLAN_EXTEND_MASK) {
-               if (dev->data->dev_conf.rxmode.hw_vlan_extend)
+               if (dev->data->dev_conf.rxmode.offloads &
+                       DEV_RX_OFFLOAD_VLAN_EXTEND)
                        PMD_INIT_LOG(ERR, "VLAN QinQ is not "
                                        "supported in fm10k");
        }
 
        if (mask & ETH_VLAN_FILTER_MASK) {
-               if (!dev->data->dev_conf.rxmode.hw_vlan_filter)
+               if (!(dev->data->dev_conf.rxmode.offloads &
+                       DEV_RX_OFFLOAD_VLAN_FILTER))
                        PMD_INIT_LOG(ERR, "VLAN filter is always on in fm10k");
        }
 
@@ -1781,6 +1799,43 @@ mempool_element_size_valid(struct rte_mempool *mp)
        return 1;
 }
 
+static uint64_t fm10k_get_rx_queue_offloads_capa(struct rte_eth_dev *dev)
+{
+       RTE_SET_USED(dev);
+
+       return (uint64_t)(DEV_RX_OFFLOAD_SCATTER);
+}
+
+static uint64_t fm10k_get_rx_port_offloads_capa(struct rte_eth_dev *dev)
+{
+       RTE_SET_USED(dev);
+
+       return  (uint64_t)(DEV_RX_OFFLOAD_VLAN_STRIP  |
+                          DEV_RX_OFFLOAD_VLAN_FILTER |
+                          DEV_RX_OFFLOAD_IPV4_CKSUM  |
+                          DEV_RX_OFFLOAD_UDP_CKSUM   |
+                          DEV_RX_OFFLOAD_TCP_CKSUM   |
+                          DEV_RX_OFFLOAD_JUMBO_FRAME |
+                          DEV_RX_OFFLOAD_CRC_STRIP   |
+                          DEV_RX_OFFLOAD_HEADER_SPLIT);
+}
+
+static int
+fm10k_check_rx_queue_offloads(struct rte_eth_dev *dev, uint64_t requested)
+{
+       uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
+       uint64_t queue_supported = fm10k_get_rx_queue_offloads_capa(dev);
+       uint64_t port_supported = fm10k_get_rx_port_offloads_capa(dev);
+
+       if ((requested & (queue_supported | port_supported)) != requested)
+               return 0;
+
+       if ((port_offloads ^ requested) & port_supported)
+               return 0;
+
+       return 1;
+}
+
 static int
 fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
        uint16_t nb_desc, unsigned int socket_id,
@@ -1794,6 +1849,18 @@ fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
 
        PMD_INIT_FUNC_TRACE();
 
+       if (!fm10k_check_rx_queue_offloads(dev, conf->offloads)) {
+               PMD_INIT_LOG(ERR, "%p: Rx queue offloads 0x%" PRIx64
+                       " don't match port offloads 0x%" PRIx64
+                       " or supported port offloads 0x%" PRIx64
+                       " or supported queue offloads 0x%" PRIx64,
+                       (void *)dev, conf->offloads,
+                       dev->data->dev_conf.rxmode.offloads,
+                       fm10k_get_rx_port_offloads_capa(dev),
+                       fm10k_get_rx_queue_offloads_capa(dev));
+               return -ENOTSUP;
+       }
+
        /* make sure the mempool element size can account for alignment. */
        if (!mempool_element_size_valid(mp)) {
                PMD_INIT_LOG(ERR, "Error : Mempool element size is too small");
@@ -1838,6 +1905,7 @@ fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
        q->queue_id = queue_id;
        q->tail_ptr = (volatile uint32_t *)
                &((uint32_t *)hw->hw_addr)[FM10K_RDT(queue_id)];
+       q->offloads = conf->offloads;
        if (handle_rxconf(q, conf))
                return -EINVAL;
 
index 498a178..b6967c0 100644 (file)
@@ -210,7 +210,7 @@ fm10k_rx_vec_condition_check(struct rte_eth_dev *dev)
 
 #ifndef RTE_FM10K_RX_OLFLAGS_ENABLE
        /* whithout rx ol_flags, no VP flag report */
-       if (rxmode->hw_vlan_extend != 0)
+       if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
                return -1;
 #endif
 
@@ -219,7 +219,7 @@ fm10k_rx_vec_condition_check(struct rte_eth_dev *dev)
                return -1;
 
        /* no header split support */
-       if (rxmode->header_split == 1)
+       if (rxmode->offloads & DEV_RX_OFFLOAD_HEADER_SPLIT)
                return -1;
 
        return 0;