liquidio CN23XX: VF xmit
authorRaghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Wed, 7 Dec 2016 16:54:35 +0000 (08:54 -0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 8 Dec 2016 18:21:39 +0000 (13:21 -0500)
Adds support for transmit functionality in VF.

Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/cavium/liquidio/cn23xx_vf_device.c
drivers/net/ethernet/cavium/liquidio/lio_vf_main.c
drivers/net/ethernet/cavium/liquidio/request_manager.c

index 108e4878e6082f7dc3051aeeca4fce84dd70f750..b6117b6a1de20d4a711a1371fa2c7558f24cd55d 100644 (file)
@@ -529,6 +529,26 @@ static u64 cn23xx_vf_msix_interrupt_handler(void *dev)
        return ret;
 }
 
+static u32 cn23xx_update_read_index(struct octeon_instr_queue *iq)
+{
+       u32 pkt_in_done = readl(iq->inst_cnt_reg);
+       u32 last_done;
+       u32 new_idx;
+
+       last_done = pkt_in_done - iq->pkt_in_done;
+       iq->pkt_in_done = pkt_in_done;
+
+       /* Modulo of the new index with the IQ size will give us
+        * the new index.  The iq->reset_instr_cnt is always zero for
+        * cn23xx, so no extra adjustments are needed.
+        */
+       new_idx = (iq->octeon_read_index +
+                  (u32)(last_done & CN23XX_PKT_IN_DONE_CNT_MASK)) %
+                 iq->max_count;
+
+       return new_idx;
+}
+
 static void cn23xx_enable_vf_interrupt(struct octeon_device *oct, u8 intr_flag)
 {
        struct octeon_cn23xx_vf *cn23xx = (struct octeon_cn23xx_vf *)oct->chip;
@@ -660,6 +680,7 @@ int cn23xx_setup_octeon_vf_device(struct octeon_device *oct)
        oct->fn_list.msix_interrupt_handler = cn23xx_vf_msix_interrupt_handler;
 
        oct->fn_list.setup_device_regs = cn23xx_setup_vf_device_regs;
+       oct->fn_list.update_iq_read_idx = cn23xx_update_read_index;
 
        oct->fn_list.enable_interrupt = cn23xx_enable_vf_interrupt;
        oct->fn_list.disable_interrupt = cn23xx_disable_vf_interrupt;
index f861a9bd6f3c3b5e7b057669d57ef3962ebbba7e..bcc8888cb5016194dc50506b224aec649b1cd673 100644 (file)
@@ -55,6 +55,21 @@ struct liquidio_if_cfg_resp {
        u64 status;
 };
 
+union tx_info {
+       u64 u64;
+       struct {
+#ifdef __BIG_ENDIAN_BITFIELD
+               u16 gso_size;
+               u16 gso_segs;
+               u32 reserved;
+#else
+               u32 reserved;
+               u16 gso_segs;
+               u16 gso_size;
+#endif
+       } s;
+};
+
 #define OCTNIC_MAX_SG  (MAX_SKB_FRAGS)
 
 #define OCTNIC_GSO_MAX_HEADER_SIZE 128
@@ -254,6 +269,19 @@ static void start_txq(struct net_device *netdev)
        }
 }
 
+/**
+ * \brief Stop a queue
+ * @param netdev network device
+ * @param q which queue to stop
+ */
+static void stop_q(struct net_device *netdev, int q)
+{
+       if (netif_is_multiqueue(netdev))
+               netif_stop_subqueue(netdev, q);
+       else
+               netif_stop_queue(netdev);
+}
+
 /**
  * Remove the node at the head of the list. The list would be empty at
  * the end of this call if there are no more nodes in the list.
@@ -944,6 +972,45 @@ static u16 select_q(struct net_device *dev, struct sk_buff *skb,
        return (u16)(qindex % (lio->linfo.num_txpciq));
 }
 
+/**
+ * \brief Setup input and output queues
+ * @param octeon_dev octeon device
+ * @param ifidx Interface index
+ *
+ * Note: Queues are with respect to the octeon device. Thus
+ * an input queue is for egress packets, and output queues
+ * are for ingress packets.
+ */
+static int setup_io_queues(struct octeon_device *octeon_dev, int ifidx)
+{
+       struct net_device *netdev;
+       int num_tx_descs;
+       struct lio *lio;
+       int retval = 0;
+       int q;
+
+       netdev = octeon_dev->props[ifidx].netdev;
+
+       lio = GET_LIO(netdev);
+
+       /* set up IQs. */
+       for (q = 0; q < lio->linfo.num_txpciq; q++) {
+               num_tx_descs = CFG_GET_NUM_TX_DESCS_NIC_IF(
+                   octeon_get_conf(octeon_dev), lio->ifidx);
+               retval = octeon_setup_iq(octeon_dev, ifidx, q,
+                                        lio->linfo.txpciq[q], num_tx_descs,
+                                        netdev_get_tx_queue(netdev, q));
+               if (retval) {
+                       dev_err(&octeon_dev->pci_dev->dev,
+                               " %s : Runtime IQ(TxQ) creation failed.\n",
+                               __func__);
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
 /**
  * \brief Net device open for LiquidIO
  * @param netdev network device
@@ -1180,6 +1247,259 @@ static int liquidio_change_mtu(struct net_device *netdev, int new_mtu)
        return 0;
 }
 
+/** \brief Transmit networks packets to the Octeon interface
+ * @param skbuff   skbuff struct to be passed to network layer.
+ * @param netdev   pointer to network device
+ * @returns whether the packet was transmitted to the device okay or not
+ *             (NETDEV_TX_OK or NETDEV_TX_BUSY)
+ */
+static int liquidio_xmit(struct sk_buff *skb, struct net_device *netdev)
+{
+       struct octnet_buf_free_info *finfo;
+       union octnic_cmd_setup cmdsetup;
+       struct octnic_data_pkt ndata;
+       struct octeon_instr_irh *irh;
+       struct oct_iq_stats *stats;
+       struct octeon_device *oct;
+       int q_idx = 0, iq_no = 0;
+       union tx_info *tx_info;
+       struct lio *lio;
+       int status = 0;
+       u64 dptr = 0;
+       u32 tag = 0;
+       int j;
+
+       lio = GET_LIO(netdev);
+       oct = lio->oct_dev;
+
+       if (netif_is_multiqueue(netdev)) {
+               q_idx = skb->queue_mapping;
+               q_idx = (q_idx % (lio->linfo.num_txpciq));
+               tag = q_idx;
+               iq_no = lio->linfo.txpciq[q_idx].s.q_no;
+       } else {
+               iq_no = lio->txq;
+       }
+
+       stats = &oct->instr_queue[iq_no]->stats;
+
+       /* Check for all conditions in which the current packet cannot be
+        * transmitted.
+        */
+       if (!(atomic_read(&lio->ifstate) & LIO_IFSTATE_RUNNING) ||
+           (!lio->linfo.link.s.link_up) || (skb->len <= 0)) {
+               netif_info(lio, tx_err, lio->netdev, "Transmit failed link_status : %d\n",
+                          lio->linfo.link.s.link_up);
+               goto lio_xmit_failed;
+       }
+
+       /* Use space in skb->cb to store info used to unmap and
+        * free the buffers.
+        */
+       finfo = (struct octnet_buf_free_info *)skb->cb;
+       finfo->lio = lio;
+       finfo->skb = skb;
+       finfo->sc = NULL;
+
+       /* Prepare the attributes for the data to be passed to OSI. */
+       memset(&ndata, 0, sizeof(struct octnic_data_pkt));
+
+       ndata.buf = finfo;
+
+       ndata.q_no = iq_no;
+
+       if (netif_is_multiqueue(netdev)) {
+               if (octnet_iq_is_full(oct, ndata.q_no)) {
+                       /* defer sending if queue is full */
+                       netif_info(lio, tx_err, lio->netdev, "Transmit failed iq:%d full\n",
+                                  ndata.q_no);
+                       stats->tx_iq_busy++;
+                       return NETDEV_TX_BUSY;
+               }
+       } else {
+               if (octnet_iq_is_full(oct, lio->txq)) {
+                       /* defer sending if queue is full */
+                       stats->tx_iq_busy++;
+                       netif_info(lio, tx_err, lio->netdev, "Transmit failed iq:%d full\n",
+                                  ndata.q_no);
+                       return NETDEV_TX_BUSY;
+               }
+       }
+
+       ndata.datasize = skb->len;
+
+       cmdsetup.u64 = 0;
+       cmdsetup.s.iq_no = iq_no;
+
+       if (skb->ip_summed == CHECKSUM_PARTIAL)
+               cmdsetup.s.transport_csum = 1;
+
+       if (!skb_shinfo(skb)->nr_frags) {
+               cmdsetup.s.u.datasize = skb->len;
+               octnet_prepare_pci_cmd(oct, &ndata.cmd, &cmdsetup, tag);
+               /* Offload checksum calculation for TCP/UDP packets */
+               dptr = dma_map_single(&oct->pci_dev->dev,
+                                     skb->data,
+                                     skb->len,
+                                     DMA_TO_DEVICE);
+               if (dma_mapping_error(&oct->pci_dev->dev, dptr)) {
+                       dev_err(&oct->pci_dev->dev, "%s DMA mapping error 1\n",
+                               __func__);
+                       return NETDEV_TX_BUSY;
+               }
+
+               ndata.cmd.cmd3.dptr = dptr;
+               finfo->dptr = dptr;
+               ndata.reqtype = REQTYPE_NORESP_NET;
+
+       } else {
+               struct skb_frag_struct *frag;
+               struct octnic_gather *g;
+               int i, frags;
+
+               spin_lock(&lio->glist_lock[q_idx]);
+               g = (struct octnic_gather *)list_delete_head(
+                   &lio->glist[q_idx]);
+               spin_unlock(&lio->glist_lock[q_idx]);
+
+               if (!g) {
+                       netif_info(lio, tx_err, lio->netdev,
+                                  "Transmit scatter gather: glist null!\n");
+                       goto lio_xmit_failed;
+               }
+
+               cmdsetup.s.gather = 1;
+               cmdsetup.s.u.gatherptrs = (skb_shinfo(skb)->nr_frags + 1);
+               octnet_prepare_pci_cmd(oct, &ndata.cmd, &cmdsetup, tag);
+
+               memset(g->sg, 0, g->sg_size);
+
+               g->sg[0].ptr[0] = dma_map_single(&oct->pci_dev->dev,
+                                                skb->data,
+                                                (skb->len - skb->data_len),
+                                                DMA_TO_DEVICE);
+               if (dma_mapping_error(&oct->pci_dev->dev, g->sg[0].ptr[0])) {
+                       dev_err(&oct->pci_dev->dev, "%s DMA mapping error 2\n",
+                               __func__);
+                       return NETDEV_TX_BUSY;
+               }
+               add_sg_size(&g->sg[0], (skb->len - skb->data_len), 0);
+
+               frags = skb_shinfo(skb)->nr_frags;
+               i = 1;
+               while (frags--) {
+                       frag = &skb_shinfo(skb)->frags[i - 1];
+
+                       g->sg[(i >> 2)].ptr[(i & 3)] =
+                               dma_map_page(&oct->pci_dev->dev,
+                                            frag->page.p,
+                                            frag->page_offset,
+                                            frag->size,
+                                            DMA_TO_DEVICE);
+                       if (dma_mapping_error(&oct->pci_dev->dev,
+                                             g->sg[i >> 2].ptr[i & 3])) {
+                               dma_unmap_single(&oct->pci_dev->dev,
+                                                g->sg[0].ptr[0],
+                                                skb->len - skb->data_len,
+                                                DMA_TO_DEVICE);
+                               for (j = 1; j < i; j++) {
+                                       frag = &skb_shinfo(skb)->frags[j - 1];
+                                       dma_unmap_page(&oct->pci_dev->dev,
+                                                      g->sg[j >> 2].ptr[j & 3],
+                                                      frag->size,
+                                                      DMA_TO_DEVICE);
+                               }
+                               dev_err(&oct->pci_dev->dev, "%s DMA mapping error 3\n",
+                                       __func__);
+                               return NETDEV_TX_BUSY;
+                       }
+
+                       add_sg_size(&g->sg[(i >> 2)], frag->size, (i & 3));
+                       i++;
+               }
+
+               dptr = dma_map_single(&oct->pci_dev->dev,
+                                     g->sg, g->sg_size,
+                                     DMA_TO_DEVICE);
+               if (dma_mapping_error(&oct->pci_dev->dev, dptr)) {
+                       dev_err(&oct->pci_dev->dev, "%s DMA mapping error 4\n",
+                               __func__);
+                       dma_unmap_single(&oct->pci_dev->dev, g->sg[0].ptr[0],
+                                        skb->len - skb->data_len,
+                                        DMA_TO_DEVICE);
+                       for (j = 1; j <= frags; j++) {
+                               frag = &skb_shinfo(skb)->frags[j - 1];
+                               dma_unmap_page(&oct->pci_dev->dev,
+                                              g->sg[j >> 2].ptr[j & 3],
+                                              frag->size, DMA_TO_DEVICE);
+                       }
+                       return NETDEV_TX_BUSY;
+               }
+
+               ndata.cmd.cmd3.dptr = dptr;
+               finfo->dptr = dptr;
+               finfo->g = g;
+
+               ndata.reqtype = REQTYPE_NORESP_NET_SG;
+       }
+
+       irh = (struct octeon_instr_irh *)&ndata.cmd.cmd3.irh;
+       tx_info = (union tx_info *)&ndata.cmd.cmd3.ossp[0];
+
+       if (skb_shinfo(skb)->gso_size) {
+               tx_info->s.gso_size = skb_shinfo(skb)->gso_size;
+               tx_info->s.gso_segs = skb_shinfo(skb)->gso_segs;
+       }
+
+       status = octnet_send_nic_data_pkt(oct, &ndata);
+       if (status == IQ_SEND_FAILED)
+               goto lio_xmit_failed;
+
+       netif_info(lio, tx_queued, lio->netdev, "Transmit queued successfully\n");
+
+       if (status == IQ_SEND_STOP) {
+               dev_err(&oct->pci_dev->dev, "Rcvd IQ_SEND_STOP signal; stopping IQ-%d\n",
+                       iq_no);
+               stop_q(lio->netdev, q_idx);
+       }
+
+       netif_trans_update(netdev);
+
+       if (skb_shinfo(skb)->gso_size)
+               stats->tx_done += skb_shinfo(skb)->gso_segs;
+       else
+               stats->tx_done++;
+       stats->tx_tot_bytes += skb->len;
+
+       return NETDEV_TX_OK;
+
+lio_xmit_failed:
+       stats->tx_dropped++;
+       netif_info(lio, tx_err, lio->netdev, "IQ%d Transmit dropped:%llu\n",
+                  iq_no, stats->tx_dropped);
+       if (dptr)
+               dma_unmap_single(&oct->pci_dev->dev, dptr,
+                                ndata.datasize, DMA_TO_DEVICE);
+       tx_buffer_free(skb);
+       return NETDEV_TX_OK;
+}
+
+/** \brief Network device Tx timeout
+ * @param netdev    pointer to network device
+ */
+static void liquidio_tx_timeout(struct net_device *netdev)
+{
+       struct lio *lio;
+
+       lio = GET_LIO(netdev);
+
+       netif_info(lio, tx_err, lio->netdev,
+                  "Transmit timeout tx_dropped:%ld, waking up queues now!!\n",
+                  netdev->stats.tx_dropped);
+       netif_trans_update(netdev);
+       txqs_wake(netdev);
+}
+
 /** Sending command to enable/disable RX checksum offload
  * @param netdev                pointer to network device
  * @param command               OCTNET_CMD_TNL_RX_CSUM_CTL
@@ -1282,8 +1602,10 @@ static int liquidio_set_features(struct net_device *netdev,
 static const struct net_device_ops lionetdevops = {
        .ndo_open               = liquidio_open,
        .ndo_stop               = liquidio_stop,
+       .ndo_start_xmit         = liquidio_xmit,
        .ndo_set_mac_address    = liquidio_set_mac,
        .ndo_set_rx_mode        = liquidio_set_mcast_list,
+       .ndo_tx_timeout         = liquidio_tx_timeout,
        .ndo_change_mtu         = liquidio_change_mtu,
        .ndo_fix_features       = liquidio_fix_features,
        .ndo_set_features       = liquidio_set_features,
@@ -1507,6 +1829,24 @@ static int setup_nic_devices(struct octeon_device *octeon_dev)
                /* Copy MAC Address to OS network device structure */
                ether_addr_copy(netdev->dev_addr, mac);
 
+               if (setup_io_queues(octeon_dev, i)) {
+                       dev_err(&octeon_dev->pci_dev->dev, "I/O queues creation failed\n");
+                       goto setup_nic_dev_fail;
+               }
+
+               /* For VFs, enable Octeon device interrupts here,
+                * as this is contingent upon IO queue setup
+                */
+               octeon_dev->fn_list.enable_interrupt(octeon_dev,
+                                                    OCTEON_ALL_INTR);
+
+               /* By default all interfaces on a single Octeon uses the same
+                * tx and rx queues
+                */
+               lio->txq = lio->linfo.txpciq[0].s.q_no;
+
+               lio->tx_qsize = octeon_get_tx_qsize(octeon_dev, lio->txq);
+
                if (setup_glists(lio, num_iqueues)) {
                        dev_err(&octeon_dev->pci_dev->dev,
                                "Gather list allocation failed\n");
index ea2b7e46631d0c1e1651dadc9b7f41bb905d024d..3ce66759e80adab406bf234926846a72828cf878 100644 (file)
@@ -394,7 +394,7 @@ lio_process_iq_request_list(struct octeon_device *oct,
                case REQTYPE_SOFT_COMMAND:
                        sc = buf;
 
-                       if (OCTEON_CN23XX_PF(oct))
+                       if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct))
                                irh = (struct octeon_instr_irh *)
                                        &sc->cmd.cmd3.irh;
                        else
@@ -607,7 +607,7 @@ octeon_prepare_soft_command(struct octeon_device *oct,
 
        oct_cfg = octeon_get_conf(oct);
 
-       if (OCTEON_CN23XX_PF(oct)) {
+       if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
                ih3 = (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
 
                ih3->pkind = oct->instr_queue[sc->iq_no]->txpciq.s.pkind;
@@ -700,7 +700,7 @@ int octeon_send_soft_command(struct octeon_device *oct,
        struct octeon_instr_irh *irh;
        u32 len;
 
-       if (OCTEON_CN23XX_PF(oct)) {
+       if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
                ih3 =  (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
                if (ih3->dlengsz) {
                        WARN_ON(!sc->dmadptr);