net/thunderx: add final bits for secondary queue support
authorKamil Rytarowski <kamil.rytarowski@caviumnetworks.com>
Fri, 30 Sep 2016 12:05:51 +0000 (14:05 +0200)
committerBruce Richardson <bruce.richardson@intel.com>
Thu, 13 Oct 2016 13:30:59 +0000 (15:30 +0200)
Summary:
 - add secondary qset support in device stats
 - add support for releasing mbufs from RBDR for >8 queues
 - add support for releasing mbufs from RX queues for >8 queues
 - support >8 queues in tx_queue_setup
 - support >8 queues in rx_queue_setup
 - support up to 96 queues per device (dev_info->max_rx_queues)
 - add secondary qset support in rbdr_rte_mempool_get
 - support >8 queues in multiprocess mode (do not reconfigure VFs)
 - setup periodic alarm accordingly for type of VFs:
   * primary VF   - handle events on queues and link status
   * secondary VF - handle events on queues
 - initialize hardware capabilities in secondary qsets

Signed-off-by: Maciej Czekaj <maciej.czekaj@caviumnetworks.com>
Signed-off-by: Kamil Rytarowski <kamil.rytarowski@caviumnetworks.com>
Signed-off-by: Zyta Szpak <zyta.szpak@semihalf.com>
Signed-off-by: Slawomir Rosek <slawomir.rosek@semihalf.com>
Signed-off-by: Radoslaw Biernacki <rad@semihalf.com>
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
drivers/net/thunderx/nicvf_ethdev.c

index 7cee99e..9925535 100644 (file)
@@ -118,7 +118,7 @@ nicvf_interrupt(void *arg)
                                nicvf_interrupt, dev);
 }
 
-static void __rte_unused
+static void
 nicvf_vf_interrupt(void *arg)
 {
        struct nicvf *nic = arg;
@@ -236,9 +236,15 @@ nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
        struct nicvf_hw_tx_qstats tx_qstats;
        struct nicvf_hw_stats port_stats;
        struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint16_t rx_start, rx_end;
+       uint16_t tx_start, tx_end;
+       size_t i;
+
+       /* RX queue indices for the first VF */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
 
        /* Reading per RX ring stats */
-       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
                if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
                        break;
 
@@ -247,8 +253,11 @@ nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
                stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
        }
 
+       /* TX queue indices for the first VF */
+       nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+
        /* Reading per TX ring stats */
-       for (qidx = 0; qidx < dev->data->nb_tx_queues; qidx++) {
+       for (qidx = tx_start; qidx <= tx_end; qidx++) {
                if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
                        break;
 
@@ -257,6 +266,40 @@ nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
                stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
        }
 
+       for (i = 0; i < nic->sqs_count; i++) {
+               struct nicvf *snic = nic->snicvf[i];
+
+               if (snic == NULL)
+                       break;
+
+               /* RX queue indices for a secondary VF */
+               nicvf_rx_range(dev, snic, &rx_start, &rx_end);
+
+               /* Reading per RX ring stats */
+               for (qidx = rx_start; qidx <= rx_end; qidx++) {
+                       if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                               break;
+
+                       nicvf_hw_get_rx_qstats(snic, &rx_qstats,
+                                              qidx % MAX_RCV_QUEUES_PER_QS);
+                       stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
+                       stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
+               }
+
+               /* TX queue indices for a secondary VF */
+               nicvf_tx_range(dev, snic, &tx_start, &tx_end);
+               /* Reading per TX ring stats */
+               for (qidx = tx_start; qidx <= tx_end; qidx++) {
+                       if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                               break;
+
+                       nicvf_hw_get_tx_qstats(snic, &tx_qstats,
+                                              qidx % MAX_SND_QUEUES_PER_QS);
+                       stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
+                       stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
+               }
+       }
+
        nicvf_hw_get_stats(nic, &port_stats);
        stats->ibytes = port_stats.rx_bytes;
        stats->ipackets = port_stats.rx_ucast_frames;
@@ -322,13 +365,36 @@ nicvf_dev_stats_reset(struct rte_eth_dev *dev)
        int i;
        uint16_t rxqs = 0, txqs = 0;
        struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint16_t rx_start, rx_end;
+       uint16_t tx_start, tx_end;
 
-       for (i = 0; i < dev->data->nb_rx_queues; i++)
+       /* Reset all primary nic counters */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+       for (i = rx_start; i <= rx_end; i++)
                rxqs |= (0x3 << (i * 2));
-       for (i = 0; i < dev->data->nb_tx_queues; i++)
+
+       nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+       for (i = tx_start; i <= tx_end; i++)
                txqs |= (0x3 << (i * 2));
 
        nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
+
+       /* Reset secondary nic queue counters */
+       for (i = 0; i < nic->sqs_count; i++) {
+               struct nicvf *snic = nic->snicvf[i];
+               if (snic == NULL)
+                       break;
+
+               nicvf_rx_range(dev, snic, &rx_start, &rx_end);
+               for (i = rx_start; i <= rx_end; i++)
+                       rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
+
+               nicvf_tx_range(dev, snic, &tx_start, &tx_end);
+               for (i = tx_start; i <= tx_end; i++)
+                       txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
+
+               nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
+       }
 }
 
 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
@@ -596,14 +662,18 @@ nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
 }
 
 static void
-nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic __rte_unused,
+nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
                        nicvf_phys_addr_t phy)
 {
        uint16_t qidx;
        void *obj;
        struct nicvf_rxq *rxq;
+       uint16_t rx_start, rx_end;
 
-       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
+       /* Get queue ranges for this VF */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
                rxq = dev->data->rx_queues[qidx];
                if (rxq->precharge_cnt) {
                        obj = (void *)nicvf_mbuff_phy2virt(phy,
@@ -861,6 +931,11 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 
        PMD_INIT_FUNC_TRACE();
 
+       if (qidx >= MAX_SND_QUEUES_PER_QS)
+               nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
+
+       qidx = qidx % MAX_SND_QUEUES_PER_QS;
+
        /* Socket id check */
        if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
                PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
@@ -895,18 +970,20 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        }
 
        /* Free memory prior to re-allocation if needed. */
-       if (dev->data->tx_queues[qidx] != NULL) {
+       if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
                PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
-                               qidx);
-               nicvf_dev_tx_queue_release(dev->data->tx_queues[qidx]);
-               dev->data->tx_queues[qidx] = NULL;
+                               nicvf_netdev_qidx(nic, qidx));
+               nicvf_dev_tx_queue_release(
+                       dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
+               dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
        }
 
        /* Allocating tx queue data structure */
        txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
                                        RTE_CACHE_LINE_SIZE, nic->node);
        if (txq == NULL) {
-               PMD_INIT_LOG(ERR, "Failed to allocate txq=%d", qidx);
+               PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
+                            nicvf_netdev_qidx(nic, qidx));
                return -ENOMEM;
        }
 
@@ -949,10 +1026,12 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        nicvf_tx_queue_reset(txq);
 
        PMD_TX_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p phys=0x%" PRIx64,
-                       qidx, txq, nb_desc, txq->desc, txq->phys);
+                       nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
+                       txq->phys);
 
-       dev->data->tx_queues[qidx] = txq;
-       dev->data->tx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
+       dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
+       dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STOPPED;
        return 0;
 }
 
@@ -967,7 +1046,8 @@ nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
        if (dev->rx_pkt_burst == NULL)
                return;
 
-       while ((rxq_cnt = nicvf_dev_rx_queue_count(dev, rxq->queue_id))) {
+       while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
+                               nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
                nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
                                        NICVF_MAX_RX_FREE_THRESH);
                PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
@@ -977,7 +1057,10 @@ nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
                }
        }
 
-       refill_cnt += nicvf_dev_rbdr_refill(dev, rxq->queue_id);
+
+       refill_cnt += nicvf_dev_rbdr_refill(dev,
+                       nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
+
        PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
                    released_pkts, refill_cnt);
 }
@@ -1149,6 +1232,11 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 
        PMD_INIT_FUNC_TRACE();
 
+       if (qidx >= MAX_RCV_QUEUES_PER_QS)
+               nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
+
+       qidx = qidx % MAX_RCV_QUEUES_PER_QS;
+
        /* Socket id check */
        if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
                PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
@@ -1191,18 +1279,20 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        }
 
        /* Free memory prior to re-allocation if needed */
-       if (dev->data->rx_queues[qidx] != NULL) {
+       if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
                PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
-                               qidx);
-               nicvf_dev_rx_queue_release(dev->data->rx_queues[qidx]);
-               dev->data->rx_queues[qidx] = NULL;
+                               nicvf_netdev_qidx(nic, qidx));
+               nicvf_dev_rx_queue_release(
+                       dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
+               dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
        }
 
        /* Allocate rxq memory */
        rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
                                        RTE_CACHE_LINE_SIZE, nic->node);
        if (rxq == NULL) {
-               PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d", qidx);
+               PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
+                            nicvf_netdev_qidx(nic, qidx));
                return -ENOMEM;
        }
 
@@ -1232,11 +1322,12 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        nicvf_rx_queue_reset(rxq);
 
        PMD_RX_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d) phy=%" PRIx64,
-                       qidx, rxq, mp->name, nb_desc,
+                       nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
                        rte_mempool_avail_count(mp), rxq->phys);
 
-       dev->data->rx_queues[qidx] = rxq;
-       dev->data->rx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
+       dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
+       dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STOPPED;
        return 0;
 }
 
@@ -1249,8 +1340,10 @@ nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 
        dev_info->min_rx_bufsize = ETHER_MIN_MTU;
        dev_info->max_rx_pktlen = NIC_HW_MAX_FRS;
-       dev_info->max_rx_queues = (uint16_t)MAX_RCV_QUEUES_PER_QS;
-       dev_info->max_tx_queues = (uint16_t)MAX_SND_QUEUES_PER_QS;
+       dev_info->max_rx_queues =
+                       (uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
+       dev_info->max_tx_queues =
+                       (uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
        dev_info->max_mac_addrs = 1;
        dev_info->max_vfs = dev->pci_dev->max_vfs;
 
@@ -1291,9 +1384,13 @@ rbdr_rte_mempool_get(void *dev, void *opaque)
        uintptr_t mbuf;
        struct nicvf_rxq *rxq;
        struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
-       struct nicvf *nic __rte_unused = (struct nicvf *)opaque;
+       struct nicvf *nic = (struct nicvf *)opaque;
+       uint16_t rx_start, rx_end;
+
+       /* Get queue ranges for this VF */
+       nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
 
-       for (qidx = 0; qidx < eth_dev->data->nb_rx_queues; qidx++) {
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
                rxq = eth_dev->data->rx_queues[qidx];
                /* Maintain equal buffer count across all pools */
                if (rxq->precharge_cnt >= rxq->qlen_mask)
@@ -1856,10 +1953,16 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
 
        /* For secondary processes, the primary has done all the work */
        if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
-               /* Setup callbacks for secondary process */
-               nicvf_set_tx_function(eth_dev);
-               nicvf_set_rx_function(eth_dev);
-               return 0;
+               if (nic) {
+                       /* Setup callbacks for secondary process */
+                       nicvf_set_tx_function(eth_dev);
+                       nicvf_set_rx_function(eth_dev);
+                       return 0;
+               } else {
+                       /* If nic == NULL than it is secondary function
+                        * so ethdev need to be released by caller */
+                       return ENOTSUP;
+               }
        }
 
        pci_dev = eth_dev->pci_dev;
@@ -1904,11 +2007,28 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
                        );
        }
 
+       ret = nicvf_base_init(nic);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
+               goto malloc_fail;
+       }
+
        if (nic->sqs_mode) {
-               PMD_INIT_LOG(INFO, "Unsupported SQS VF detected, Detaching...");
-               /* Detach port by returning Positive error number */
-               ret = ENOTSUP;
-               goto alarm_fail;
+               /* Push nic to stack of secondary vfs */
+               nicvf_svf_push(nic);
+
+               /* Steal nic pointer from the device for further reuse */
+               eth_dev->data->dev_private = NULL;
+
+               nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
+               ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "Failed to start period alarm");
+                       goto fail;
+               }
+
+               /* Detach port by returning postive error number */
+               return ENOTSUP;
        }
 
        eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
@@ -1929,12 +2049,6 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
                goto malloc_fail;
        }
 
-       ret = nicvf_base_init(nic);
-       if (ret) {
-               PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
-               goto malloc_fail;
-       }
-
        PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
                eth_dev->data->port_id, nic->vendor_id, nic->device_id,
                nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],