liquidio:Scatter gather list per IQ
authorRaghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Tue, 14 Jun 2016 23:54:45 +0000 (16:54 -0700)
committerDavid S. Miller <davem@davemloft.net>
Thu, 16 Jun 2016 04:44:31 +0000 (21:44 -0700)
This patch is to allocate and manage scatter gather lists per
input queue(iq's) and remove queue's interdependence.

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: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/cavium/liquidio/lio_main.c
drivers/net/ethernet/cavium/liquidio/octeon_network.h

index 3477a3c15b5672679df5fd1c7942e5bc27e97944..0daa89a78d7c57493aba654c1a6c969ab9fc3a57 100644 (file)
@@ -166,6 +166,8 @@ struct octnic_gather {
         *  received from the IP layer.
         */
        struct octeon_sg_entry *sg;
+
+       u64 sg_dma_ptr;
 };
 
 /** This structure is used by NIC driver to store information required
@@ -791,64 +793,116 @@ static inline struct list_head *list_delete_head(struct list_head *root)
 }
 
 /**
- * \brief Delete gather list
+ * \brief Delete gather lists
  * @param lio per-network private data
  */
-static void delete_glist(struct lio *lio)
+static void delete_glists(struct lio *lio)
 {
        struct octnic_gather *g;
+       int i;
 
-       do {
-               g = (struct octnic_gather *)
-                   list_delete_head(&lio->glist);
-               if (g) {
-                       if (g->sg)
-                               kfree((void *)((unsigned long)g->sg -
-                                               g->adjust));
-                       kfree(g);
-               }
-       } while (g);
+       if (!lio->glist)
+               return;
+
+       for (i = 0; i < lio->linfo.num_txpciq; i++) {
+               do {
+                       g = (struct octnic_gather *)
+                               list_delete_head(&lio->glist[i]);
+                       if (g) {
+                               if (g->sg) {
+                                       dma_unmap_single(&lio->oct_dev->
+                                                        pci_dev->dev,
+                                                        g->sg_dma_ptr,
+                                                        g->sg_size,
+                                                        DMA_TO_DEVICE);
+                                       kfree((void *)((unsigned long)g->sg -
+                                                      g->adjust));
+                               }
+                               kfree(g);
+                       }
+               } while (g);
+       }
+
+       kfree((void *)lio->glist);
 }
 
 /**
- * \brief Setup gather list
+ * \brief Setup gather lists
  * @param lio per-network private data
  */
-static int setup_glist(struct lio *lio)
+static int setup_glists(struct octeon_device *oct, struct lio *lio, int num_iqs)
 {
-       int i;
+       int i, j;
        struct octnic_gather *g;
 
-       INIT_LIST_HEAD(&lio->glist);
+       lio->glist_lock = kcalloc(num_iqs, sizeof(*lio->glist_lock),
+                                 GFP_KERNEL);
+       if (!lio->glist_lock)
+               return 1;
 
-       for (i = 0; i < lio->tx_qsize; i++) {
-               g = kzalloc(sizeof(*g), GFP_KERNEL);
-               if (!g)
-                       break;
+       lio->glist = kcalloc(num_iqs, sizeof(*lio->glist),
+                            GFP_KERNEL);
+       if (!lio->glist) {
+               kfree((void *)lio->glist_lock);
+               return 1;
+       }
 
-               g->sg_size =
-                       ((ROUNDUP4(OCTNIC_MAX_SG) >> 2) * OCT_SG_ENTRY_SIZE);
+       for (i = 0; i < num_iqs; i++) {
+               int numa_node = cpu_to_node(i % num_online_cpus());
 
-               g->sg = kmalloc(g->sg_size + 8, GFP_KERNEL);
-               if (!g->sg) {
-                       kfree(g);
-                       break;
+               spin_lock_init(&lio->glist_lock[i]);
+
+               INIT_LIST_HEAD(&lio->glist[i]);
+
+               for (j = 0; j < lio->tx_qsize; j++) {
+                       g = kzalloc_node(sizeof(*g), GFP_KERNEL,
+                                        numa_node);
+                       if (!g)
+                               g = kzalloc(sizeof(*g), GFP_KERNEL);
+                       if (!g)
+                               break;
+
+                       g->sg_size = ((ROUNDUP4(OCTNIC_MAX_SG) >> 2) *
+                                     OCT_SG_ENTRY_SIZE);
+
+                       g->sg = kmalloc_node(g->sg_size + 8,
+                                            GFP_KERNEL, numa_node);
+                       if (!g->sg)
+                               g->sg = kmalloc(g->sg_size + 8, GFP_KERNEL);
+                       if (!g->sg) {
+                               kfree(g);
+                               break;
+                       }
+
+                       /* The gather component should be aligned on 64-bit
+                        * boundary
+                        */
+                       if (((unsigned long)g->sg) & 7) {
+                               g->adjust = 8 - (((unsigned long)g->sg) & 7);
+                               g->sg = (struct octeon_sg_entry *)
+                                       ((unsigned long)g->sg + g->adjust);
+                       }
+                       g->sg_dma_ptr = dma_map_single(&oct->pci_dev->dev,
+                                                      g->sg, g->sg_size,
+                                                      DMA_TO_DEVICE);
+                       if (dma_mapping_error(&oct->pci_dev->dev,
+                                             g->sg_dma_ptr)) {
+                               kfree((void *)((unsigned long)g->sg -
+                                              g->adjust));
+                               kfree(g);
+                               break;
+                       }
+
+                       list_add_tail(&g->list, &lio->glist[i]);
                }
 
-               /* The gather component should be aligned on 64-bit boundary */
-               if (((unsigned long)g->sg) & 7) {
-                       g->adjust = 8 - (((unsigned long)g->sg) & 7);
-                       g->sg = (struct octeon_sg_entry *)
-                               ((unsigned long)g->sg + g->adjust);
+               if (j != lio->tx_qsize) {
+                       delete_glists(lio);
+                       return 1;
                }
-               list_add_tail(&g->list, &lio->glist);
        }
 
-       if (i == lio->tx_qsize)
-               return 0;
-
-       delete_glist(lio);
-       return 1;
+       return 0;
 }
 
 /**
@@ -1209,7 +1263,7 @@ static void liquidio_destroy_nic_device(struct octeon_device *oct, int ifidx)
        if (atomic_read(&lio->ifstate) & LIO_IFSTATE_REGISTERED)
                unregister_netdev(netdev);
 
-       delete_glist(lio);
+       delete_glists(lio);
 
        free_netdev(netdev);
 
@@ -1331,6 +1385,16 @@ static int octeon_pci_os_setup(struct octeon_device *oct)
        return 0;
 }
 
+static inline int skb_iq(struct lio *lio, struct sk_buff *skb)
+{
+       int q = 0;
+
+       if (netif_is_multiqueue(lio->netdev))
+               q = skb->queue_mapping % lio->linfo.num_txpciq;
+
+       return q;
+}
+
 /**
  * \brief Check Tx queue state for a given network buffer
  * @param lio per-network private data
@@ -1388,7 +1452,7 @@ static void free_netsgbuf(void *buf)
        struct sk_buff *skb;
        struct lio *lio;
        struct octnic_gather *g;
-       int i, frags;
+       int i, frags, iq;
 
        finfo = (struct octnet_buf_free_info *)buf;
        skb = finfo->skb;
@@ -1410,13 +1474,13 @@ static void free_netsgbuf(void *buf)
                i++;
        }
 
-       dma_unmap_single(&lio->oct_dev->pci_dev->dev,
-                        finfo->dptr, g->sg_size,
-                        DMA_TO_DEVICE);
+       dma_sync_single_for_cpu(&lio->oct_dev->pci_dev->dev,
+                               g->sg_dma_ptr, g->sg_size, DMA_TO_DEVICE);
 
-       spin_lock(&lio->lock);
-       list_add_tail(&g->list, &lio->glist);
-       spin_unlock(&lio->lock);
+       iq = skb_iq(lio, skb);
+       spin_lock(&lio->glist_lock[iq]);
+       list_add_tail(&g->list, &lio->glist[iq]);
+       spin_unlock(&lio->glist_lock[iq]);
 
        check_txq_state(lio, skb);     /* mq support: sub-queue state check */
 
@@ -1434,7 +1498,7 @@ static void free_netsgbuf_with_resp(void *buf)
        struct sk_buff *skb;
        struct lio *lio;
        struct octnic_gather *g;
-       int i, frags;
+       int i, frags, iq;
 
        sc = (struct octeon_soft_command *)buf;
        skb = (struct sk_buff *)sc->callback_arg;
@@ -1458,13 +1522,14 @@ static void free_netsgbuf_with_resp(void *buf)
                i++;
        }
 
-       dma_unmap_single(&lio->oct_dev->pci_dev->dev,
-                        finfo->dptr, g->sg_size,
-                        DMA_TO_DEVICE);
+       dma_sync_single_for_cpu(&lio->oct_dev->pci_dev->dev,
+                               g->sg_dma_ptr, g->sg_size, DMA_TO_DEVICE);
 
-       spin_lock(&lio->lock);
-       list_add_tail(&g->list, &lio->glist);
-       spin_unlock(&lio->lock);
+       iq = skb_iq(lio, skb);
+
+       spin_lock(&lio->glist_lock[iq]);
+       list_add_tail(&g->list, &lio->glist[iq]);
+       spin_unlock(&lio->glist_lock[iq]);
 
        /* Don't free the skb yet */
 
@@ -2683,7 +2748,8 @@ static int liquidio_xmit(struct sk_buff *skb, struct net_device *netdev)
        struct oct_iq_stats *stats;
        int status = 0;
        int q_idx = 0, iq_no = 0;
-       int xmit_more;
+       int xmit_more, j;
+       u64 dptr = 0;
        u32 tag = 0;
 
        lio = GET_LIO(netdev);
@@ -2826,9 +2892,10 @@ static int liquidio_xmit(struct sk_buff *skb, struct net_device *netdev)
                struct skb_frag_struct *frag;
                struct octnic_gather *g;
 
-               spin_lock(&lio->lock);
-               g = (struct octnic_gather *)list_delete_head(&lio->glist);
-               spin_unlock(&lio->lock);
+               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,
@@ -2865,21 +2932,31 @@ static int liquidio_xmit(struct sk_buff *skb, struct net_device *netdev)
                                             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++;
                }
 
-               ndata.cmd.dptr = dma_map_single(&oct->pci_dev->dev,
-                                               g->sg, g->sg_size,
-                                               DMA_TO_DEVICE);
-               if (dma_mapping_error(&oct->pci_dev->dev, ndata.cmd.dptr)) {
-                       dev_err(&oct->pci_dev->dev, "%s DMA mapping error 3\n",
-                               __func__);
-                       dma_unmap_single(&oct->pci_dev->dev, g->sg[0].ptr[0],
-                                        skb->len - skb->data_len,
-                                        DMA_TO_DEVICE);
-                       return NETDEV_TX_BUSY;
-               }
+               dma_sync_single_for_device(&oct->pci_dev->dev, g->sg_dma_ptr,
+                                          g->sg_size, DMA_TO_DEVICE);
+               dptr = g->sg_dma_ptr;
 
                finfo->dptr = ndata.cmd.dptr;
                finfo->g = g;
@@ -3301,7 +3378,6 @@ static int setup_nic_devices(struct octeon_device *octeon_dev)
                lio->oct_dev = octeon_dev;
                lio->octprops = props;
                lio->netdev = netdev;
-               spin_lock_init(&lio->lock);
 
                dev_dbg(&octeon_dev->pci_dev->dev,
                        "if%d gmx: %d hw_addr: 0x%llx\n", i,
@@ -3331,7 +3407,7 @@ static int setup_nic_devices(struct octeon_device *octeon_dev)
                lio->tx_qsize = octeon_get_tx_qsize(octeon_dev, lio->txq);
                lio->rx_qsize = octeon_get_rx_qsize(octeon_dev, lio->rxq);
 
-               if (setup_glist(lio)) {
+               if (setup_glists(octeon_dev, lio, num_iqueues)) {
                        dev_err(&octeon_dev->pci_dev->dev,
                                "Gather list allocation failed\n");
                        goto setup_nic_dev_fail;
index b3abe5818fd3a7267da02bdb9850354d2d6f0e74..0a50bacb982397badcce174fde9e8bd0f2ab298d 100644 (file)
@@ -48,11 +48,11 @@ struct lio {
         */
        int rxq;
 
-       /** Guards the glist */
-       spinlock_t lock;
+       /** Guards each glist */
+       spinlock_t *glist_lock;
 
-       /** Linked list of gather components */
-       struct list_head glist;
+       /** Array of gather component linked lists */
+       struct list_head *glist;
 
        /** Pointer to the NIC properties for the Octeon device this network
         *  interface is associated with.