net/thunderx: remove private data to ethdev link
authorKamil Rytarowski <kamil.rytarowski@caviumnetworks.com>
Fri, 30 Sep 2016 12:05:46 +0000 (14:05 +0200)
committerBruce Richardson <bruce.richardson@intel.com>
Thu, 13 Oct 2016 13:30:59 +0000 (15:30 +0200)
In case of the multiprocess mode a shared nicvf struct between processes
cannot point with the eth_dev pointer to master device, therefore remove it
along with references to it refactoring the code where needed.

This change fixes multiprocess issues detected in stats.

Fixes: 7413feee662d ("net/thunderx: add device start/stop and close")

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
drivers/net/thunderx/nicvf_rxtx.c
drivers/net/thunderx/nicvf_struct.h

index aa99eda..385ca48 100644 (file)
@@ -497,14 +497,14 @@ nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
 }
 
 static int
-nicvf_qset_cq_alloc(struct nicvf *nic, struct nicvf_rxq *rxq, uint16_t qidx,
-                   uint32_t desc_cnt)
+nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
+                   struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
 {
        const struct rte_memzone *rz;
        uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
 
-       rz = rte_eth_dma_zone_reserve(nic->eth_dev, "cq_ring", qidx, ring_size,
-                                       NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
+       rz = rte_eth_dma_zone_reserve(dev, "cq_ring", qidx, ring_size,
+                                     NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
        if (rz == NULL) {
                PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
                return -ENOMEM;
@@ -520,13 +520,13 @@ nicvf_qset_cq_alloc(struct nicvf *nic, struct nicvf_rxq *rxq, uint16_t qidx,
 }
 
 static int
-nicvf_qset_sq_alloc(struct nicvf *nic,  struct nicvf_txq *sq, uint16_t qidx,
-                   uint32_t desc_cnt)
+nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
+                   struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
 {
        const struct rte_memzone *rz;
        uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
 
-       rz = rte_eth_dma_zone_reserve(nic->eth_dev, "sq", qidx, ring_size,
+       rz = rte_eth_dma_zone_reserve(dev, "sq", qidx, ring_size,
                                NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
        if (rz == NULL) {
                PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
@@ -543,7 +543,8 @@ nicvf_qset_sq_alloc(struct nicvf *nic,  struct nicvf_txq *sq, uint16_t qidx,
 }
 
 static int
-nicvf_qset_rbdr_alloc(struct nicvf *nic, uint32_t desc_cnt, uint32_t buffsz)
+nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
+                     uint32_t desc_cnt, uint32_t buffsz)
 {
        struct nicvf_rbdr *rbdr;
        const struct rte_memzone *rz;
@@ -558,7 +559,7 @@ nicvf_qset_rbdr_alloc(struct nicvf *nic, uint32_t desc_cnt, uint32_t buffsz)
        }
 
        ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
-       rz = rte_eth_dma_zone_reserve(nic->eth_dev, "rbdr", 0, ring_size,
+       rz = rte_eth_dma_zone_reserve(dev, "rbdr", 0, ring_size,
                                   NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
        if (rz == NULL) {
                PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
@@ -583,14 +584,15 @@ nicvf_qset_rbdr_alloc(struct nicvf *nic, uint32_t desc_cnt, uint32_t buffsz)
 }
 
 static void
-nicvf_rbdr_release_mbuf(struct nicvf *nic, nicvf_phys_addr_t phy)
+nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic __rte_unused,
+                       nicvf_phys_addr_t phy)
 {
        uint16_t qidx;
        void *obj;
        struct nicvf_rxq *rxq;
 
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_rx_queues; qidx++) {
-               rxq = nic->eth_dev->data->rx_queues[qidx];
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
+               rxq = dev->data->rx_queues[qidx];
                if (rxq->precharge_cnt) {
                        obj = (void *)nicvf_mbuff_phy2virt(phy,
                                                           rxq->mbuf_phys_off);
@@ -602,7 +604,7 @@ nicvf_rbdr_release_mbuf(struct nicvf *nic, nicvf_phys_addr_t phy)
 }
 
 static inline void
-nicvf_rbdr_release_mbufs(struct nicvf *nic)
+nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
 {
        uint32_t qlen_mask, head;
        struct rbdr_entry_t *entry;
@@ -612,7 +614,7 @@ nicvf_rbdr_release_mbufs(struct nicvf *nic)
        head = rbdr->head;
        while (head != rbdr->tail) {
                entry = rbdr->desc + head;
-               nicvf_rbdr_release_mbuf(nic, entry->full_addr);
+               nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
                head++;
                head = head & qlen_mask;
        }
@@ -724,14 +726,13 @@ nicvf_configure_rss(struct rte_eth_dev *dev)
                        dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
        PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
                    dev->data->dev_conf.rxmode.mq_mode,
-                   nic->eth_dev->data->nb_rx_queues,
-                   nic->eth_dev->data->dev_conf.lpbk_mode, rsshf);
+                   dev->data->nb_rx_queues,
+                   dev->data->dev_conf.lpbk_mode, rsshf);
 
        if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
                ret = nicvf_rss_term(nic);
        else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
-               ret = nicvf_rss_config(nic,
-                                      nic->eth_dev->data->nb_rx_queues, rsshf);
+               ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
        if (ret)
                PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
 
@@ -915,7 +916,7 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
                return -ENOMEM;
        }
 
-       if (nicvf_qset_sq_alloc(nic, txq, qidx, nb_desc)) {
+       if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
                PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
                nicvf_dev_tx_queue_release(txq);
                return -ENOMEM;
@@ -932,12 +933,11 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 }
 
 static inline void
-nicvf_rx_queue_release_mbufs(struct nicvf_rxq *rxq)
+nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
 {
        uint32_t rxq_cnt;
        uint32_t nb_pkts, released_pkts = 0;
        uint32_t refill_cnt = 0;
-       struct rte_eth_dev *dev = rxq->nic->eth_dev;
        struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
 
        if (dev->rx_pkt_burst == NULL)
@@ -1017,7 +1017,7 @@ nicvf_stop_rx_queue(struct rte_eth_dev *dev, uint16_t qidx)
 
        other_error = ret;
        rxq = dev->data->rx_queues[qidx];
-       nicvf_rx_queue_release_mbufs(rxq);
+       nicvf_rx_queue_release_mbufs(dev, rxq);
        nicvf_rx_queue_reset(rxq);
 
        ret = nicvf_qset_cq_reclaim(nic, qidx);
@@ -1163,7 +1163,7 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 
 
        /* Alloc completion queue */
-       if (nicvf_qset_cq_alloc(nic, rxq, rxq->queue_id, nb_desc)) {
+       if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
                PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
                nicvf_dev_rx_queue_release(rxq);
                return -ENOMEM;
@@ -1281,7 +1281,7 @@ nicvf_dev_start(struct rte_eth_dev *dev)
         */
 
        /* Validate RBDR buff size */
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_rx_queues; qidx++) {
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
                rxq = dev->data->rx_queues[qidx];
                mbp_priv = rte_mempool_get_priv(rxq->pool);
                buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
@@ -1299,7 +1299,7 @@ nicvf_dev_start(struct rte_eth_dev *dev)
        }
 
        /* Validate mempool attributes */
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_rx_queues; qidx++) {
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
                rxq = dev->data->rx_queues[qidx];
                rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
                mbuf = rte_pktmbuf_alloc(rxq->pool);
@@ -1323,12 +1323,12 @@ nicvf_dev_start(struct rte_eth_dev *dev)
 
        /* Check the level of buffers in the pool */
        total_rxq_desc = 0;
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_rx_queues; qidx++) {
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
                rxq = dev->data->rx_queues[qidx];
                /* Count total numbers of rxq descs */
                total_rxq_desc += rxq->qlen_mask + 1;
                exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
-               exp_buffs *= nic->eth_dev->data->nb_rx_queues;
+               exp_buffs *= dev->data->nb_rx_queues;
                if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
                        PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
                                     rxq->pool->name,
@@ -1354,7 +1354,7 @@ nicvf_dev_start(struct rte_eth_dev *dev)
 
        /* Allocate RBDR and RBDR ring desc */
        nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
-       ret = nicvf_qset_rbdr_alloc(nic, nb_rbdr_desc, rbdrsz);
+       ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
        if (ret) {
                PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc");
                goto qset_reclaim;
@@ -1379,7 +1379,7 @@ nicvf_dev_start(struct rte_eth_dev *dev)
                     nic->rbdr->tail, nb_rbdr_desc);
 
        /* Configure RX queues */
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_rx_queues; qidx++) {
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
                ret = nicvf_start_rx_queue(dev, qidx);
                if (ret)
                        goto start_rxq_error;
@@ -1389,7 +1389,7 @@ nicvf_dev_start(struct rte_eth_dev *dev)
        nicvf_vlan_hw_strip(nic, dev->data->dev_conf.rxmode.hw_vlan_strip);
 
        /* Configure TX queues */
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_tx_queues; qidx++) {
+       for (qidx = 0; qidx < dev->data->nb_tx_queues; qidx++) {
                ret = nicvf_start_tx_queue(dev, qidx);
                if (ret)
                        goto start_txq_error;
@@ -1448,14 +1448,14 @@ nicvf_dev_start(struct rte_eth_dev *dev)
 qset_rss_error:
        nicvf_rss_term(nic);
 start_txq_error:
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_tx_queues; qidx++)
+       for (qidx = 0; qidx < dev->data->nb_tx_queues; qidx++)
                nicvf_stop_tx_queue(dev, qidx);
 start_rxq_error:
-       for (qidx = 0; qidx < nic->eth_dev->data->nb_rx_queues; qidx++)
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++)
                nicvf_stop_rx_queue(dev, qidx);
 qset_rbdr_reclaim:
        nicvf_qset_rbdr_reclaim(nic, 0);
-       nicvf_rbdr_release_mbufs(nic);
+       nicvf_rbdr_release_mbufs(dev, nic);
 qset_rbdr_free:
        if (nic->rbdr) {
                rte_free(nic->rbdr);
@@ -1501,7 +1501,7 @@ nicvf_dev_stop(struct rte_eth_dev *dev)
 
        /* Move all charged buffers in RBDR back to pool */
        if (nic->rbdr != NULL)
-               nicvf_rbdr_release_mbufs(nic);
+               nicvf_rbdr_release_mbufs(dev, nic);
 
        /* Reclaim CPI configuration */
        if (!nic->sqs_mode) {
@@ -1666,7 +1666,6 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
        nic->vendor_id = pci_dev->id.vendor_id;
        nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
        nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
-       nic->eth_dev = eth_dev;
 
        PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
                        pci_dev->id.vendor_id, pci_dev->id.device_id,
index e15c730..fc43b74 100644 (file)
@@ -368,7 +368,8 @@ nicvf_fill_rbdr(struct nicvf_rxq *rxq, int to_fill)
        void *obj_p[NICVF_MAX_RX_FREE_THRESH] __rte_cache_aligned;
 
        if (unlikely(rte_mempool_get_bulk(rxq->pool, obj_p, to_fill) < 0)) {
-               rxq->nic->eth_dev->data->rx_mbuf_alloc_failed += to_fill;
+               rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
+                       to_fill;
                return 0;
        }
 
index a72f752..c900e12 100644 (file)
@@ -113,7 +113,6 @@ struct nicvf {
        uint16_t subsystem_vendor_id;
        struct nicvf_rbdr *rbdr;
        struct nicvf_rss_reta_info rss_info;
-       struct rte_eth_dev *eth_dev;
        struct rte_intr_handle intr_handle;
        uint8_t cpi_alg;
        uint16_t mtu;