net/mlx4: Add data structures to support N-Ports per VF
authorMatan Barak <matanb@mellanox.com>
Wed, 19 Mar 2014 16:11:50 +0000 (18:11 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 20 Mar 2014 20:18:29 +0000 (16:18 -0400)
Adds the required data structures to support VFs with N (1 or 2)
ports instead of always using the number of physical ports.

Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx4/main.c
include/linux/mlx4/device.h

index abd0b1d277aaee2778a7103b799999e3b0cf5c41..e1a55857af712c2fbce3e542ef96a1877e6f9e41 100644 (file)
@@ -2193,6 +2193,10 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
        struct mlx4_dev *dev;
        int err;
        int port;
+       int nvfs[MLX4_MAX_PORTS + 1], prb_vf[MLX4_MAX_PORTS + 1];
+       unsigned total_vfs = 0;
+       int sriov_initialized = 0;
+       unsigned int i;
 
        pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
 
@@ -2207,17 +2211,39 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
         * per port, we must limit the number of VFs to 63 (since their are
         * 128 MACs)
         */
-       if (num_vfs >= MLX4_MAX_NUM_VF) {
+       for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]);
+            total_vfs += nvfs[i], i++) {
+               nvfs[i] = i == MLX4_MAX_PORTS ? num_vfs : 0;
+               if (nvfs[i] < 0) {
+                       dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
+                       return -EINVAL;
+               }
+       }
+       for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]); i++) {
+               prb_vf[i] = i == MLX4_MAX_PORTS ? probe_vf : 0;
+               if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
+                       dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
+                       return -EINVAL;
+               }
+       }
+       if (total_vfs >= MLX4_MAX_NUM_VF) {
                dev_err(&pdev->dev,
                        "Requested more VF's (%d) than allowed (%d)\n",
-                       num_vfs, MLX4_MAX_NUM_VF - 1);
+                       total_vfs, MLX4_MAX_NUM_VF - 1);
                return -EINVAL;
        }
 
-       if (num_vfs < 0) {
-               pr_err("num_vfs module parameter cannot be negative\n");
-               return -EINVAL;
+       for (i = 0; i < MLX4_MAX_PORTS; i++) {
+               if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
+                       dev_err(&pdev->dev,
+                               "Requested more VF's (%d) for port (%d) than allowed (%d)\n",
+                               nvfs[i] + nvfs[2], i + 1,
+                               MLX4_MAX_NUM_VF_P_PORT - 1);
+                       return -EINVAL;
+               }
        }
+
+
        /*
         * Check for BARs.
         */
@@ -2292,11 +2318,23 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
        if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
                /* When acting as pf, we normally skip vfs unless explicitly
                 * requested to probe them. */
-               if (num_vfs && extended_func_num(pdev) > probe_vf) {
-                       mlx4_warn(dev, "Skipping virtual function:%d\n",
-                                               extended_func_num(pdev));
-                       err = -ENODEV;
-                       goto err_free_dev;
+               if (total_vfs) {
+                       unsigned vfs_offset = 0;
+                       for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
+                            vfs_offset + nvfs[i] < extended_func_num(pdev);
+                            vfs_offset += nvfs[i], i++)
+                               ;
+                       if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
+                               err = -ENODEV;
+                               goto err_free_dev;
+                       }
+                       if ((extended_func_num(pdev) - vfs_offset)
+                           > prb_vf[i]) {
+                               mlx4_warn(dev, "Skipping virtual function:%d\n",
+                                         extended_func_num(pdev));
+                               err = -ENODEV;
+                               goto err_free_dev;
+                       }
                }
                mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
                dev->flags |= MLX4_FLAG_SLAVE;
@@ -2316,22 +2354,30 @@ static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
                        }
                }
 
-               if (num_vfs) {
-                       mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", num_vfs);
-
-                       atomic_inc(&pf_loading);
-                       err = pci_enable_sriov(pdev, num_vfs);
-                       atomic_dec(&pf_loading);
-
-                       if (err) {
-                               mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n",
-                                        err);
+               if (total_vfs) {
+                       mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n",
+                                 total_vfs);
+                       dev->dev_vfs = kzalloc(
+                                       total_vfs * sizeof(*dev->dev_vfs),
+                                       GFP_KERNEL);
+                       if (NULL == dev->dev_vfs) {
+                               mlx4_err(dev, "Failed to allocate memory for VFs\n");
                                err = 0;
                        } else {
-                               mlx4_warn(dev, "Running in master mode\n");
-                               dev->flags |= MLX4_FLAG_SRIOV |
-                                             MLX4_FLAG_MASTER;
-                               dev->num_vfs = num_vfs;
+                               atomic_inc(&pf_loading);
+                               err = pci_enable_sriov(pdev, total_vfs);
+                               atomic_dec(&pf_loading);
+                               if (err) {
+                                       mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n",
+                                                err);
+                                       err = 0;
+                               } else {
+                                       mlx4_warn(dev, "Running in master mode\n");
+                                       dev->flags |= MLX4_FLAG_SRIOV |
+                                                     MLX4_FLAG_MASTER;
+                                       dev->num_vfs = total_vfs;
+                                       sriov_initialized = 1;
+                               }
                        }
                }
 
@@ -2396,12 +2442,24 @@ slave_start:
        /* In master functions, the communication channel must be initialized
         * after obtaining its address from fw */
        if (mlx4_is_master(dev)) {
+               unsigned sum = 0;
                err = mlx4_multi_func_init(dev);
                if (err) {
                        mlx4_err(dev, "Failed to init master mfunc"
                                 "interface, aborting.\n");
                        goto err_close;
                }
+               if (sriov_initialized) {
+                       for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]); i++) {
+                               unsigned j;
+                               for (j = 0; j < nvfs[i]; ++sum, ++j) {
+                                       dev->dev_vfs[sum].min_port =
+                                               i < 2 ? i + 1 : 1;
+                                       dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
+                                               dev->caps.num_ports;
+                               }
+                       }
+               }
        }
 
        err = mlx4_alloc_eq_table(dev);
@@ -2509,6 +2567,8 @@ err_rel_own:
        if (!mlx4_is_slave(dev))
                mlx4_free_ownership(dev);
 
+       kfree(priv->dev.dev_vfs);
+
 err_free_dev:
        kfree(priv);
 
@@ -2595,6 +2655,7 @@ static void mlx4_remove_one(struct pci_dev *pdev)
                kfree(dev->caps.qp0_proxy);
                kfree(dev->caps.qp1_tunnel);
                kfree(dev->caps.qp1_proxy);
+               kfree(dev->dev_vfs);
 
                kfree(priv);
                pci_release_regions(pdev);
index f211b51dc726bf06daef5c16697d79bee7737a35..b3044f32aef1e1adc42ade7cfc8c58b3315f2d2c 100644 (file)
@@ -84,6 +84,7 @@ enum {
 enum {
        MLX4_MAX_NUM_PF         = 16,
        MLX4_MAX_NUM_VF         = 64,
+       MLX4_MAX_NUM_VF_P_PORT  = 64,
        MLX4_MFUNC_MAX          = 80,
        MLX4_MAX_EQ_NUM         = 1024,
        MLX4_MFUNC_EQ_NUM       = 4,
@@ -664,6 +665,11 @@ struct mlx4_quotas {
        int xrcd;
 };
 
+struct mlx4_vf_dev {
+       u8                      min_port;
+       u8                      n_ports;
+};
+
 struct mlx4_dev {
        struct pci_dev         *pdev;
        unsigned long           flags;
@@ -679,6 +685,7 @@ struct mlx4_dev {
        int                     oper_log_mgm_entry_size;
        u64                     regid_promisc_array[MLX4_MAX_PORTS + 1];
        u64                     regid_allmulti_array[MLX4_MAX_PORTS + 1];
+       struct mlx4_vf_dev     *dev_vfs;
 };
 
 struct mlx4_eqe {