net/sfc: use switch port ID as representor ID
[dpdk.git] / drivers / net / sfc / sfc_ethdev.c
index 6b83722..0297e79 100644 (file)
@@ -28,7 +28,9 @@
 #include "sfc_flow.h"
 #include "sfc_dp.h"
 #include "sfc_dp_rx.h"
+#include "sfc_repr.h"
 #include "sfc_sw_stats.h"
+#include "sfc_switch.h"
 
 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
 #define SFC_XSTAT_ID_INVALID_NAME '\0'
@@ -344,6 +346,8 @@ sfc_dev_close(struct rte_eth_dev *dev)
                return 0;
        }
 
+       sfc_pre_detach(sa);
+
        sfc_adapter_lock(sa);
        switch (sa->state) {
        case SFC_ETHDEV_STARTED:
@@ -1915,6 +1919,177 @@ sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
        return sap->dp_rx->intr_disable(rxq_info->dp);
 }
 
+struct sfc_mport_journal_ctx {
+       struct sfc_adapter              *sa;
+       uint16_t                        switch_domain_id;
+       uint32_t                        mcdi_handle;
+       bool                            controllers_assigned;
+       efx_pcie_interface_t            *controllers;
+       size_t                          nb_controllers;
+};
+
+static int
+sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
+                              efx_pcie_interface_t intf)
+{
+       efx_pcie_interface_t *new_controllers;
+       size_t i, target;
+       size_t new_size;
+
+       if (ctx->controllers == NULL) {
+               ctx->controllers = rte_malloc("sfc_controller_mapping",
+                                             sizeof(ctx->controllers[0]), 0);
+               if (ctx->controllers == NULL)
+                       return ENOMEM;
+
+               ctx->controllers[0] = intf;
+               ctx->nb_controllers = 1;
+
+               return 0;
+       }
+
+       for (i = 0; i < ctx->nb_controllers; i++) {
+               if (ctx->controllers[i] == intf)
+                       return 0;
+               if (ctx->controllers[i] > intf)
+                       break;
+       }
+       target = i;
+
+       ctx->nb_controllers += 1;
+       new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
+
+       new_controllers = rte_realloc(ctx->controllers, new_size, 0);
+       if (new_controllers == NULL) {
+               rte_free(ctx->controllers);
+               return ENOMEM;
+       }
+       ctx->controllers = new_controllers;
+
+       for (i = target + 1; i < ctx->nb_controllers; i++)
+               ctx->controllers[i] = ctx->controllers[i - 1];
+
+       ctx->controllers[target] = intf;
+
+       return 0;
+}
+
+static efx_rc_t
+sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
+                               efx_mport_desc_t *mport)
+{
+       efx_mport_sel_t ethdev_mport;
+       int rc;
+
+       sfc_dbg(ctx->sa,
+               "processing mport id %u (controller %u pf %u vf %u)",
+               mport->emd_id.id, mport->emd_vnic.ev_intf,
+               mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
+       efx_mae_mport_invalid(&ethdev_mport);
+
+       if (!ctx->controllers_assigned) {
+               rc = sfc_journal_ctx_add_controller(ctx,
+                                                   mport->emd_vnic.ev_intf);
+               if (rc != 0)
+                       return rc;
+       }
+
+       return 0;
+}
+
+static efx_rc_t
+sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
+                            size_t mport_len)
+{
+       struct sfc_mport_journal_ctx *ctx = data;
+
+       if (ctx == NULL || ctx->sa == NULL) {
+               sfc_err(ctx->sa, "received NULL context or SFC adapter");
+               return EINVAL;
+       }
+
+       if (mport_len != sizeof(*mport)) {
+               sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
+               return EINVAL;
+       }
+
+       SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
+
+       /*
+        * If a zombie flag is set, it means the mport has been marked for
+        * deletion and cannot be used for any new operations. The mport will
+        * be destroyed completely once all references to it are released.
+        */
+       if (mport->emd_zombie) {
+               sfc_dbg(ctx->sa, "mport is a zombie, skipping");
+               return 0;
+       }
+       if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
+               sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
+               return 0;
+       }
+       if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
+               sfc_dbg(ctx->sa, "mport is not a function, skipping");
+               return 0;
+       }
+       if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
+               sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
+               return 0;
+       }
+
+       return sfc_process_mport_journal_entry(ctx, mport);
+}
+
+static int
+sfc_process_mport_journal(struct sfc_adapter *sa)
+{
+       struct sfc_mport_journal_ctx ctx;
+       const efx_pcie_interface_t *controllers;
+       size_t nb_controllers;
+       efx_rc_t efx_rc;
+       int rc;
+
+       memset(&ctx, 0, sizeof(ctx));
+       ctx.sa = sa;
+       ctx.switch_domain_id = sa->mae.switch_domain_id;
+
+       efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
+       if (efx_rc != 0) {
+               sfc_err(sa, "failed to get own MCDI handle");
+               SFC_ASSERT(efx_rc > 0);
+               return efx_rc;
+       }
+
+       rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
+                                              &controllers, &nb_controllers);
+       if (rc != 0) {
+               sfc_err(sa, "failed to get controller mapping");
+               return rc;
+       }
+
+       ctx.controllers_assigned = controllers != NULL;
+       ctx.controllers = NULL;
+       ctx.nb_controllers = 0;
+
+       efx_rc = efx_mae_read_mport_journal(sa->nic,
+                                           sfc_process_mport_journal_cb, &ctx);
+       if (efx_rc != 0) {
+               sfc_err(sa, "failed to process MAE mport journal");
+               SFC_ASSERT(efx_rc > 0);
+               return efx_rc;
+       }
+
+       if (controllers == NULL) {
+               rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
+                                                          ctx.controllers,
+                                                          ctx.nb_controllers);
+               if (rc != 0)
+                       return rc;
+       }
+
+       return 0;
+}
+
 static const struct eth_dev_ops sfc_eth_dev_ops = {
        .dev_configure                  = sfc_dev_configure,
        .dev_start                      = sfc_dev_start,
@@ -1964,6 +2139,10 @@ static const struct eth_dev_ops sfc_eth_dev_ops = {
        .pool_ops_supported             = sfc_pool_ops_supported,
 };
 
+struct sfc_ethdev_init_data {
+       uint16_t                nb_representors;
+};
+
 /**
  * Duplicate a string in potentially shared memory required for
  * multi-process support.
@@ -2243,7 +2422,7 @@ sfc_register_dp(void)
 }
 
 static int
-sfc_parse_switch_mode(struct sfc_adapter *sa)
+sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
 {
        const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
        const char *switch_mode = NULL;
@@ -2258,7 +2437,8 @@ sfc_parse_switch_mode(struct sfc_adapter *sa)
 
        if (switch_mode == NULL) {
                sa->switchdev = encp->enc_mae_supported &&
-                               !encp->enc_datapath_cap_evb;
+                               (!encp->enc_datapath_cap_evb ||
+                                has_representors);
        } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
                sa->switchdev = false;
        } else if (strcasecmp(switch_mode,
@@ -2283,10 +2463,11 @@ fail_kvargs:
 }
 
 static int
-sfc_eth_dev_init(struct rte_eth_dev *dev)
+sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
 {
        struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
        struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct sfc_ethdev_init_data *init_data = init_params;
        uint32_t logtype_main;
        struct sfc_adapter *sa;
        int rc;
@@ -2377,7 +2558,7 @@ sfc_eth_dev_init(struct rte_eth_dev *dev)
         * Selecting a default switch mode requires the NIC to be probed and
         * to have its capabilities filled in.
         */
-       rc = sfc_parse_switch_mode(sa);
+       rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
        if (rc != 0)
                goto fail_switch_mode;
 
@@ -2462,11 +2643,297 @@ static const struct rte_pci_id pci_id_sfc_efx_map[] = {
        { .vendor_id = 0 /* sentinel */ }
 };
 
+static int
+sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
+{
+       struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
+       int rc;
+
+       if (args != NULL) {
+               rc = rte_eth_devargs_parse(args, &eth_da);
+               if (rc != 0) {
+                       SFC_GENERIC_LOG(ERR,
+                                       "Failed to parse generic devargs '%s'",
+                                       args);
+                       return rc;
+               }
+       }
+
+       *devargs = eth_da;
+
+       return 0;
+}
+
+static int
+sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
+                          struct sfc_ethdev_init_data *init_data,
+                          struct rte_eth_dev **devp,
+                          bool *dev_created)
+{
+       struct rte_eth_dev *dev;
+       bool created = false;
+       int rc;
+
+       dev = rte_eth_dev_allocated(pci_dev->device.name);
+       if (dev == NULL) {
+               rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
+                                       sizeof(struct sfc_adapter_shared),
+                                       eth_dev_pci_specific_init, pci_dev,
+                                       sfc_eth_dev_init, init_data);
+               if (rc != 0) {
+                       SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
+                                       pci_dev->device.name);
+                       return rc;
+               }
+
+               created = true;
+
+               dev = rte_eth_dev_allocated(pci_dev->device.name);
+               if (dev == NULL) {
+                       SFC_GENERIC_LOG(ERR,
+                               "Failed to find allocated sfc ethdev '%s'",
+                               pci_dev->device.name);
+                       return -ENODEV;
+               }
+       }
+
+       *devp = dev;
+       *dev_created = created;
+
+       return 0;
+}
+
+static int
+sfc_eth_dev_create_repr(struct sfc_adapter *sa,
+                       efx_pcie_interface_t controller,
+                       uint16_t port,
+                       uint16_t repr_port,
+                       enum rte_eth_representor_type type)
+{
+       struct sfc_repr_entity_info entity;
+       efx_mport_sel_t mport_sel;
+       int rc;
+
+       switch (type) {
+       case RTE_ETH_REPRESENTOR_NONE:
+               return 0;
+       case RTE_ETH_REPRESENTOR_VF:
+       case RTE_ETH_REPRESENTOR_PF:
+               break;
+       case RTE_ETH_REPRESENTOR_SF:
+               sfc_err(sa, "SF representors are not supported");
+               return ENOTSUP;
+       default:
+               sfc_err(sa, "unknown representor type: %d", type);
+               return ENOTSUP;
+       }
+
+       rc = efx_mae_mport_by_pcie_mh_function(controller,
+                                              port,
+                                              repr_port,
+                                              &mport_sel);
+       if (rc != 0) {
+               sfc_err(sa,
+                       "failed to get m-port selector for controller %u port %u repr_port %u: %s",
+                       controller, port, repr_port, rte_strerror(-rc));
+               return rc;
+       }
+
+       memset(&entity, 0, sizeof(entity));
+       entity.type = type;
+       entity.intf = controller;
+       entity.pf = port;
+       entity.vf = repr_port;
+
+       rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
+                            &mport_sel);
+       if (rc != 0) {
+               sfc_err(sa,
+                       "failed to create representor for controller %u port %u repr_port %u: %s",
+                       controller, port, repr_port, rte_strerror(-rc));
+               return rc;
+       }
+
+       return 0;
+}
+
+static int
+sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
+                            const struct rte_eth_devargs *eth_da,
+                            efx_pcie_interface_t controller,
+                            uint16_t port)
+{
+       int first_error = 0;
+       uint16_t i;
+       int rc;
+
+       if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
+               return sfc_eth_dev_create_repr(sa, controller, port,
+                                              EFX_PCI_VF_INVALID,
+                                              eth_da->type);
+       }
+
+       for (i = 0; i < eth_da->nb_representor_ports; i++) {
+               rc = sfc_eth_dev_create_repr(sa, controller, port,
+                                            eth_da->representor_ports[i],
+                                            eth_da->type);
+               if (rc != 0 && first_error == 0)
+                       first_error = rc;
+       }
+
+       return first_error;
+}
+
+static int
+sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
+                                  const struct rte_eth_devargs *eth_da,
+                                  efx_pcie_interface_t controller)
+{
+       const efx_nic_cfg_t *encp;
+       int first_error = 0;
+       uint16_t default_port;
+       uint16_t i;
+       int rc;
+
+       if (eth_da->nb_ports == 0) {
+               encp = efx_nic_cfg_get(sa->nic);
+               default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
+               return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
+                                                   default_port);
+       }
+
+       for (i = 0; i < eth_da->nb_ports; i++) {
+               rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
+                                                 eth_da->ports[i]);
+               if (rc != 0 && first_error == 0)
+                       first_error = rc;
+       }
+
+       return first_error;
+}
+
+static int
+sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
+                               const struct rte_eth_devargs *eth_da)
+{
+       efx_pcie_interface_t intf;
+       const efx_nic_cfg_t *encp;
+       struct sfc_adapter *sa;
+       uint16_t switch_domain_id;
+       uint16_t i;
+       int rc;
+
+       sa = sfc_adapter_by_eth_dev(dev);
+       switch_domain_id = sa->mae.switch_domain_id;
+
+       switch (eth_da->type) {
+       case RTE_ETH_REPRESENTOR_NONE:
+               return 0;
+       case RTE_ETH_REPRESENTOR_PF:
+       case RTE_ETH_REPRESENTOR_VF:
+               break;
+       case RTE_ETH_REPRESENTOR_SF:
+               sfc_err(sa, "SF representors are not supported");
+               return -ENOTSUP;
+       default:
+               sfc_err(sa, "unknown representor type: %d",
+                       eth_da->type);
+               return -ENOTSUP;
+       }
+
+       if (!sa->switchdev) {
+               sfc_err(sa, "cannot create representors in non-switchdev mode");
+               return -EINVAL;
+       }
+
+       if (!sfc_repr_available(sfc_sa2shared(sa))) {
+               sfc_err(sa, "cannot create representors: unsupported");
+
+               return -ENOTSUP;
+       }
+
+       /*
+        * This is needed to construct the DPDK controller -> EFX interface
+        * mapping.
+        */
+       sfc_adapter_lock(sa);
+       rc = sfc_process_mport_journal(sa);
+       sfc_adapter_unlock(sa);
+       if (rc != 0) {
+               SFC_ASSERT(rc > 0);
+               return -rc;
+       }
+
+       if (eth_da->nb_mh_controllers > 0) {
+               for (i = 0; i < eth_da->nb_mh_controllers; i++) {
+                       rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
+                                               eth_da->mh_controllers[i],
+                                               &intf);
+                       if (rc != 0) {
+                               sfc_err(sa, "failed to get representor");
+                               continue;
+                       }
+                       sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
+               }
+       } else {
+               encp = efx_nic_cfg_get(sa->nic);
+               sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
+       }
+
+       return 0;
+}
+
 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
        struct rte_pci_device *pci_dev)
 {
-       return rte_eth_dev_pci_generic_probe(pci_dev,
-               sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
+       struct sfc_ethdev_init_data init_data;
+       struct rte_eth_devargs eth_da;
+       struct rte_eth_dev *dev;
+       bool dev_created;
+       int rc;
+
+       if (pci_dev->device.devargs != NULL) {
+               rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
+                                          &eth_da);
+               if (rc != 0)
+                       return rc;
+       } else {
+               memset(&eth_da, 0, sizeof(eth_da));
+       }
+
+       /* If no VF representors specified, check for PF ones */
+       if (eth_da.nb_representor_ports > 0)
+               init_data.nb_representors = eth_da.nb_representor_ports;
+       else
+               init_data.nb_representors = eth_da.nb_ports;
+
+       if (init_data.nb_representors > 0 &&
+           rte_eal_process_type() != RTE_PROC_PRIMARY) {
+               SFC_GENERIC_LOG(ERR,
+                       "Create representors from secondary process not supported, dev '%s'",
+                       pci_dev->device.name);
+               return -ENOTSUP;
+       }
+
+       /*
+        * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
+        * if it does not already exist. Re-probing an existing device is
+        * expected to allow additional representors to be configured.
+        */
+       rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
+                                       &dev_created);
+       if (rc != 0)
+               return rc;
+
+       rc = sfc_eth_dev_create_representors(dev, &eth_da);
+       if (rc != 0) {
+               if (dev_created)
+                       (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
+
+               return rc;
+       }
+
+       return 0;
 }
 
 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
@@ -2478,7 +2945,8 @@ static struct rte_pci_driver sfc_efx_pmd = {
        .id_table = pci_id_sfc_efx_map,
        .drv_flags =
                RTE_PCI_DRV_INTR_LSC |
-               RTE_PCI_DRV_NEED_MAPPING,
+               RTE_PCI_DRV_NEED_MAPPING |
+               RTE_PCI_DRV_PROBE_AGAIN,
        .probe = sfc_eth_dev_pci_probe,
        .remove = sfc_eth_dev_pci_remove,
 };