ethdev: move info filling of PCI into drivers
[dpdk.git] / drivers / net / ena / ena_ethdev.c
index e2716b5..020c02a 100644 (file)
@@ -38,6 +38,7 @@
 #include <rte_dev.h>
 #include <rte_errno.h>
 #include <rte_version.h>
+#include <rte_eal_memconfig.h>
 
 #include "ena_ethdev.h"
 #include "ena_logs.h"
@@ -168,10 +169,9 @@ static const struct ena_stats ena_stats_ena_com_strings[] = {
 #define PCI_DEVICE_ID_ENA_LLQ_VF       0xEC21
 
 static struct rte_pci_id pci_id_ena_map[] = {
-#define RTE_PCI_DEV_ID_DECL_ENA(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
-       RTE_PCI_DEV_ID_DECL_ENA(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF)
-       RTE_PCI_DEV_ID_DECL_ENA(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF)
-       {.device_id = 0},
+       { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
+       { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
+       { .device_id = 0 },
 };
 
 static int ena_device_init(struct ena_com_dev *ena_dev,
@@ -232,6 +232,18 @@ static struct eth_dev_ops ena_dev_ops = {
        .reta_query           = ena_rss_reta_query,
 };
 
+#define NUMA_NO_NODE   SOCKET_ID_ANY
+
+static inline int ena_cpu_to_node(int cpu)
+{
+       struct rte_config *config = rte_eal_get_configuration();
+
+       if (likely(cpu < RTE_MAX_MEMZONE))
+               return config->mem_config->memzone[cpu].socket_id;
+
+       return NUMA_NO_NODE;
+}
+
 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
                                       struct ena_com_rx_ctx *ena_rx_ctx)
 {
@@ -330,15 +342,18 @@ static void ena_config_host_info(struct ena_com_dev *ena_dev)
 
        host_info->os_type = ENA_ADMIN_OS_DPDK;
        host_info->kernel_ver = RTE_VERSION;
-       strncpy((char *)host_info->kernel_ver_str, rte_version(),
-               strlen(rte_version()));
+       snprintf((char *)host_info->kernel_ver_str,
+                sizeof(host_info->kernel_ver_str),
+                "%s", rte_version());
        host_info->os_dist = RTE_VERSION;
-       strncpy((char *)host_info->os_dist_str, rte_version(),
-               strlen(rte_version()));
+       snprintf((char *)host_info->os_dist_str,
+                sizeof(host_info->os_dist_str),
+                "%s", rte_version());
        host_info->driver_version =
                (DRV_MODULE_VER_MAJOR) |
                (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
-               (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
+               (DRV_MODULE_VER_SUBMINOR <<
+                       ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
 
        rc = ena_com_set_host_attributes(ena_dev);
        if (rc) {
@@ -661,8 +676,7 @@ static void ena_rx_queue_release_bufs(struct ena_ring *ring)
                if (m)
                        __rte_mbuf_raw_free(m);
 
-               ring->next_to_clean =
-                       ENA_CIRC_INC(ring->next_to_clean, 1, ring->ring_size);
+               ring->next_to_clean++;
        }
 }
 
@@ -677,8 +691,7 @@ static void ena_tx_queue_release_bufs(struct ena_ring *ring)
                if (tx_buf->mbuf)
                        rte_pktmbuf_free(tx_buf->mbuf);
 
-               ring->next_to_clean =
-                       ENA_CIRC_INC(ring->next_to_clean, 1, ring->ring_size);
+               ring->next_to_clean++;
        }
 }
 
@@ -911,8 +924,8 @@ static int ena_queue_restart(struct ena_ring *ring)
        if (ring->type == ENA_RING_TYPE_TX)
                return 0;
 
-       rc = ena_populate_rx_queue(ring, ring->ring_size - 1);
-       if ((unsigned int)rc != ring->ring_size - 1) {
+       rc = ena_populate_rx_queue(ring, ring->ring_size);
+       if ((unsigned int)rc != ring->ring_size) {
                PMD_INIT_LOG(ERR, "Failed to populate rx ring !\n");
                return (-1);
        }
@@ -947,6 +960,13 @@ static int ena_tx_queue_setup(struct rte_eth_dev *dev,
                return -1;
        }
 
+       if (!rte_is_power_of_2(nb_desc)) {
+               RTE_LOG(ERR, PMD,
+                       "Unsupported size of RX queue: %d is not a power of 2.",
+                       nb_desc);
+               return -EINVAL;
+       }
+
        if (nb_desc > adapter->tx_ring_size) {
                RTE_LOG(ERR, PMD,
                        "Unsupported size of TX queue (max size: %d)\n",
@@ -961,6 +981,7 @@ static int ena_tx_queue_setup(struct rte_eth_dev *dev,
        ctx.msix_vector = -1; /* admin interrupts not used */
        ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
        ctx.queue_size = adapter->tx_ring_size;
+       ctx.numa_node = ena_cpu_to_node(queue_idx);
 
        rc = ena_com_create_io_queue(ena_dev, &ctx);
        if (rc) {
@@ -1040,6 +1061,13 @@ static int ena_rx_queue_setup(struct rte_eth_dev *dev,
                return -1;
        }
 
+       if (!rte_is_power_of_2(nb_desc)) {
+               RTE_LOG(ERR, PMD,
+                       "Unsupported size of TX queue: %d is not a power of 2.",
+                       nb_desc);
+               return -EINVAL;
+       }
+
        if (nb_desc > adapter->rx_ring_size) {
                RTE_LOG(ERR, PMD,
                        "Unsupported size of RX queue (max size: %d)\n",
@@ -1054,6 +1082,7 @@ static int ena_rx_queue_setup(struct rte_eth_dev *dev,
        ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
        ctx.msix_vector = -1; /* admin interrupts not used */
        ctx.queue_size = adapter->rx_ring_size;
+       ctx.numa_node = ena_cpu_to_node(queue_idx);
 
        rc = ena_com_create_io_queue(ena_dev, &ctx);
        if (rc)
@@ -1098,23 +1127,25 @@ static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
 {
        unsigned int i;
        int rc;
-       unsigned int ring_size = rxq->ring_size;
-       unsigned int ring_mask = ring_size - 1;
-       int next_to_use = rxq->next_to_use & ring_mask;
+       uint16_t ring_size = rxq->ring_size;
+       uint16_t ring_mask = ring_size - 1;
+       uint16_t next_to_use = rxq->next_to_use;
+       uint16_t in_use;
        struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
 
        if (unlikely(!count))
                return 0;
 
-       ena_assert_msg((((ENA_CIRC_COUNT(rxq->next_to_use, rxq->next_to_clean,
-                                        rxq->ring_size)) +
-                        count) < rxq->ring_size), "bad ring state");
+       in_use = rxq->next_to_use - rxq->next_to_clean;
+       ena_assert_msg(((in_use + count) <= ring_size), "bad ring state");
 
-       count = RTE_MIN(count, ring_size - next_to_use);
+       count = RTE_MIN(count,
+                       (uint16_t)(ring_size - (next_to_use & ring_mask)));
 
        /* get resources for incoming packets */
        rc = rte_mempool_get_bulk(rxq->mb_pool,
-                                 (void **)(&mbufs[next_to_use]), count);
+                                 (void **)(&mbufs[next_to_use & ring_mask]),
+                                 count);
        if (unlikely(rc < 0)) {
                rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
                PMD_RX_LOG(DEBUG, "there are no enough free buffers");
@@ -1122,7 +1153,8 @@ static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
        }
 
        for (i = 0; i < count; i++) {
-               struct rte_mbuf *mbuf = mbufs[next_to_use];
+               uint16_t next_to_use_masked = next_to_use & ring_mask;
+               struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
                struct ena_com_buf ebuf;
 
                rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
@@ -1131,18 +1163,22 @@ static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
                ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
                /* pass resource to device */
                rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
-                                               &ebuf, next_to_use);
+                                               &ebuf, next_to_use_masked);
                if (unlikely(rc)) {
                        RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
                        break;
                }
-               next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use, ring_size);
+               next_to_use++;
        }
 
-       rte_wmb();
-       rxq->next_to_use = next_to_use;
-       /* let HW know that it can fill buffers with data */
-       ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
+       /* When we submitted free recources to device... */
+       if (i > 0) {
+               /* ...let HW know that it can fill buffers with data */
+               rte_wmb();
+               ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
+
+               rxq->next_to_use = next_to_use;
+       }
 
        return i;
 }
@@ -1151,6 +1187,7 @@ static int ena_device_init(struct ena_com_dev *ena_dev,
                           struct ena_com_dev_get_features_ctx *get_feat_ctx)
 {
        int rc;
+       bool readless_supported;
 
        /* Initialize mmio registers */
        rc = ena_com_mmio_reg_read_request_init(ena_dev);
@@ -1159,6 +1196,14 @@ static int ena_device_init(struct ena_com_dev *ena_dev,
                return rc;
        }
 
+       /* The PCIe configuration space revision id indicate if mmio reg
+        * read is disabled.
+        */
+       readless_supported =
+               !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
+                              & ENA_MMIO_DISABLE_REG_READ);
+       ena_com_set_mmio_read_mode(ena_dev, readless_supported);
+
        /* reset device */
        rc = ena_com_dev_reset(ena_dev);
        if (rc) {
@@ -1391,6 +1436,8 @@ static void ena_infos_get(struct rte_eth_dev *dev,
        ena_dev = &adapter->ena_dev;
        ena_assert_msg(ena_dev != NULL, "Uninitialized device");
 
+       dev_info->pci_dev = dev->pci_dev;
+
        dev_info->speed_capa =
                        ETH_LINK_SPEED_1G   |
                        ETH_LINK_SPEED_2_5G |
@@ -1445,7 +1492,7 @@ static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        unsigned int ring_size = rx_ring->ring_size;
        unsigned int ring_mask = ring_size - 1;
        uint16_t next_to_clean = rx_ring->next_to_clean;
-       int desc_in_use = 0;
+       uint16_t desc_in_use = 0;
        unsigned int recv_idx = 0;
        struct rte_mbuf *mbuf = NULL;
        struct rte_mbuf *mbuf_head = NULL;
@@ -1463,8 +1510,7 @@ static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                return 0;
        }
 
-       desc_in_use = ENA_CIRC_COUNT(rx_ring->next_to_use,
-                                    next_to_clean, ring_size);
+       desc_in_use = rx_ring->next_to_use - next_to_clean;
        if (unlikely(nb_pkts > desc_in_use))
                nb_pkts = desc_in_use;
 
@@ -1505,8 +1551,7 @@ static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 
                        mbuf_prev = mbuf;
                        segments++;
-                       next_to_clean =
-                               ENA_RX_RING_IDX_NEXT(next_to_clean, ring_size);
+                       next_to_clean++;
                }
 
                /* fill mbuf attributes if any */
@@ -1519,10 +1564,10 @@ static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        }
 
        /* Burst refill to save doorbells, memory barriers, const interval */
-       if (ring_size - desc_in_use - 1 > ENA_RING_DESCS_RATIO(ring_size))
-               ena_populate_rx_queue(rx_ring, ring_size - desc_in_use - 1);
+       if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
+               ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
 
-       rx_ring->next_to_clean = next_to_clean & ring_mask;
+       rx_ring->next_to_clean = next_to_clean;
 
        return recv_idx;
 }
@@ -1531,7 +1576,8 @@ static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                                  uint16_t nb_pkts)
 {
        struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
-       unsigned int next_to_use = tx_ring->next_to_use;
+       uint16_t next_to_use = tx_ring->next_to_use;
+       uint16_t next_to_clean = tx_ring->next_to_clean;
        struct rte_mbuf *mbuf;
        unsigned int ring_size = tx_ring->ring_size;
        unsigned int ring_mask = ring_size - 1;
@@ -1539,7 +1585,7 @@ static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        struct ena_tx_buffer *tx_info;
        struct ena_com_buf *ebuf;
        uint16_t rc, req_id, total_tx_descs = 0;
-       int sent_idx = 0;
+       uint16_t sent_idx = 0, empty_tx_reqs;
        int nb_hw_desc;
 
        /* Check adapter state */
@@ -1549,10 +1595,14 @@ static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                return 0;
        }
 
+       empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
+       if (nb_pkts > empty_tx_reqs)
+               nb_pkts = empty_tx_reqs;
+
        for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
                mbuf = tx_pkts[sent_idx];
 
-               req_id = tx_ring->empty_tx_reqs[next_to_use];
+               req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
                tx_info = &tx_ring->tx_buffer_info[req_id];
                tx_info->mbuf = mbuf;
                tx_info->num_of_bufs = 0;
@@ -1615,12 +1665,17 @@ static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 
                tx_info->tx_descs = nb_hw_desc;
 
-               next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use, ring_size);
+               next_to_use++;
        }
 
-       /* Let HW do it's best :-) */
-       rte_wmb();
-       ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
+       /* If there are ready packets to be xmitted... */
+       if (sent_idx > 0) {
+               /* ...let HW do its best :-) */
+               rte_wmb();
+               ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
+
+               tx_ring->next_to_use = next_to_use;
+       }
 
        /* Clear complete packets  */
        while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
@@ -1633,44 +1688,34 @@ static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                rte_pktmbuf_free(mbuf);
 
                /* Put back descriptor to the ring for reuse */
-               tx_ring->empty_tx_reqs[tx_ring->next_to_clean] = req_id;
-               tx_ring->next_to_clean =
-                       ENA_TX_RING_IDX_NEXT(tx_ring->next_to_clean,
-                                            tx_ring->ring_size);
+               tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
+               next_to_clean++;
 
                /* If too many descs to clean, leave it for another run */
                if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
                        break;
        }
 
-       /* acknowledge completion of sent packets */
-       ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
-       tx_ring->next_to_use = next_to_use;
+       if (total_tx_descs > 0) {
+               /* acknowledge completion of sent packets */
+               ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
+               tx_ring->next_to_clean = next_to_clean;
+       }
+
        return sent_idx;
 }
 
 static struct eth_driver rte_ena_pmd = {
-       {
-               .name = "rte_ena_pmd",
+       .pci_drv = {
                .id_table = pci_id_ena_map,
                .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
+               .probe = rte_eth_dev_pci_probe,
+               .remove = rte_eth_dev_pci_remove,
        },
        .eth_dev_init = eth_ena_dev_init,
        .dev_private_size = sizeof(struct ena_adapter),
 };
 
-static int
-rte_ena_pmd_init(const char *name __rte_unused,
-                const char *params __rte_unused)
-{
-       rte_eth_driver_register(&rte_ena_pmd);
-       return 0;
-};
-
-struct rte_driver ena_pmd_drv = {
-       .type = PMD_PDEV,
-       .init = rte_ena_pmd_init,
-};
-
-PMD_REGISTER_DRIVER(ena_pmd_drv, ena);
-DRIVER_REGISTER_PCI_TABLE(ena, pci_id_ena_map);
+RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd.pci_drv);
+RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
+RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio");