ethdev: introduce available Rx descriptors threshold
[dpdk.git] / lib / ethdev / ethdev_driver.c
index fb7323f..a285f21 100644 (file)
@@ -2,7 +2,632 @@
  * Copyright(c) 2022 Intel Corporation
  */
 
+#include <rte_kvargs.h>
+#include <rte_malloc.h>
+
 #include "ethdev_driver.h"
+#include "ethdev_private.h"
+
+/**
+ * A set of values to describe the possible states of a switch domain.
+ */
+enum rte_eth_switch_domain_state {
+       RTE_ETH_SWITCH_DOMAIN_UNUSED = 0,
+       RTE_ETH_SWITCH_DOMAIN_ALLOCATED
+};
+
+/**
+ * Array of switch domains available for allocation. Array is sized to
+ * RTE_MAX_ETHPORTS elements as there cannot be more active switch domains than
+ * ethdev ports in a single process.
+ */
+static struct rte_eth_dev_switch {
+       enum rte_eth_switch_domain_state state;
+} eth_dev_switch_domains[RTE_MAX_ETHPORTS];
+
+static struct rte_eth_dev *
+eth_dev_allocated(const char *name)
+{
+       uint16_t i;
+
+       RTE_BUILD_BUG_ON(RTE_MAX_ETHPORTS >= UINT16_MAX);
+
+       for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
+               if (rte_eth_devices[i].data != NULL &&
+                   strcmp(rte_eth_devices[i].data->name, name) == 0)
+                       return &rte_eth_devices[i];
+       }
+       return NULL;
+}
+
+static uint16_t
+eth_dev_find_free_port(void)
+{
+       uint16_t i;
+
+       for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
+               /* Using shared name field to find a free port. */
+               if (eth_dev_shared_data->data[i].name[0] == '\0') {
+                       RTE_ASSERT(rte_eth_devices[i].state ==
+                                  RTE_ETH_DEV_UNUSED);
+                       return i;
+               }
+       }
+       return RTE_MAX_ETHPORTS;
+}
+
+static struct rte_eth_dev *
+eth_dev_get(uint16_t port_id)
+{
+       struct rte_eth_dev *eth_dev = &rte_eth_devices[port_id];
+
+       eth_dev->data = &eth_dev_shared_data->data[port_id];
+
+       return eth_dev;
+}
+
+struct rte_eth_dev *
+rte_eth_dev_allocate(const char *name)
+{
+       uint16_t port_id;
+       struct rte_eth_dev *eth_dev = NULL;
+       size_t name_len;
+
+       name_len = strnlen(name, RTE_ETH_NAME_MAX_LEN);
+       if (name_len == 0) {
+               RTE_ETHDEV_LOG(ERR, "Zero length Ethernet device name\n");
+               return NULL;
+       }
+
+       if (name_len >= RTE_ETH_NAME_MAX_LEN) {
+               RTE_ETHDEV_LOG(ERR, "Ethernet device name is too long\n");
+               return NULL;
+       }
+
+       eth_dev_shared_data_prepare();
+
+       /* Synchronize port creation between primary and secondary threads. */
+       rte_spinlock_lock(&eth_dev_shared_data->ownership_lock);
+
+       if (eth_dev_allocated(name) != NULL) {
+               RTE_ETHDEV_LOG(ERR,
+                       "Ethernet device with name %s already allocated\n",
+                       name);
+               goto unlock;
+       }
+
+       port_id = eth_dev_find_free_port();
+       if (port_id == RTE_MAX_ETHPORTS) {
+               RTE_ETHDEV_LOG(ERR,
+                       "Reached maximum number of Ethernet ports\n");
+               goto unlock;
+       }
+
+       eth_dev = eth_dev_get(port_id);
+       strlcpy(eth_dev->data->name, name, sizeof(eth_dev->data->name));
+       eth_dev->data->port_id = port_id;
+       eth_dev->data->backer_port_id = RTE_MAX_ETHPORTS;
+       eth_dev->data->mtu = RTE_ETHER_MTU;
+       pthread_mutex_init(&eth_dev->data->flow_ops_mutex, NULL);
+
+unlock:
+       rte_spinlock_unlock(&eth_dev_shared_data->ownership_lock);
+
+       return eth_dev;
+}
+
+struct rte_eth_dev *
+rte_eth_dev_allocated(const char *name)
+{
+       struct rte_eth_dev *ethdev;
+
+       eth_dev_shared_data_prepare();
+
+       rte_spinlock_lock(&eth_dev_shared_data->ownership_lock);
+
+       ethdev = eth_dev_allocated(name);
+
+       rte_spinlock_unlock(&eth_dev_shared_data->ownership_lock);
+
+       return ethdev;
+}
+
+/*
+ * Attach to a port already registered by the primary process, which
+ * makes sure that the same device would have the same port ID both
+ * in the primary and secondary process.
+ */
+struct rte_eth_dev *
+rte_eth_dev_attach_secondary(const char *name)
+{
+       uint16_t i;
+       struct rte_eth_dev *eth_dev = NULL;
+
+       eth_dev_shared_data_prepare();
+
+       /* Synchronize port attachment to primary port creation and release. */
+       rte_spinlock_lock(&eth_dev_shared_data->ownership_lock);
+
+       for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
+               if (strcmp(eth_dev_shared_data->data[i].name, name) == 0)
+                       break;
+       }
+       if (i == RTE_MAX_ETHPORTS) {
+               RTE_ETHDEV_LOG(ERR,
+                       "Device %s is not driven by the primary process\n",
+                       name);
+       } else {
+               eth_dev = eth_dev_get(i);
+               RTE_ASSERT(eth_dev->data->port_id == i);
+       }
+
+       rte_spinlock_unlock(&eth_dev_shared_data->ownership_lock);
+       return eth_dev;
+}
+
+int
+rte_eth_dev_callback_process(struct rte_eth_dev *dev,
+       enum rte_eth_event_type event, void *ret_param)
+{
+       struct rte_eth_dev_callback *cb_lst;
+       struct rte_eth_dev_callback dev_cb;
+       int rc = 0;
+
+       rte_spinlock_lock(&eth_dev_cb_lock);
+       TAILQ_FOREACH(cb_lst, &(dev->link_intr_cbs), next) {
+               if (cb_lst->cb_fn == NULL || cb_lst->event != event)
+                       continue;
+               dev_cb = *cb_lst;
+               cb_lst->active = 1;
+               if (ret_param != NULL)
+                       dev_cb.ret_param = ret_param;
+
+               rte_spinlock_unlock(&eth_dev_cb_lock);
+               rc = dev_cb.cb_fn(dev->data->port_id, dev_cb.event,
+                               dev_cb.cb_arg, dev_cb.ret_param);
+               rte_spinlock_lock(&eth_dev_cb_lock);
+               cb_lst->active = 0;
+       }
+       rte_spinlock_unlock(&eth_dev_cb_lock);
+       return rc;
+}
+
+void
+rte_eth_dev_probing_finish(struct rte_eth_dev *dev)
+{
+       if (dev == NULL)
+               return;
+
+       /*
+        * for secondary process, at that point we expect device
+        * to be already 'usable', so shared data and all function pointers
+        * for fast-path devops have to be setup properly inside rte_eth_dev.
+        */
+       if (rte_eal_process_type() == RTE_PROC_SECONDARY)
+               eth_dev_fp_ops_setup(rte_eth_fp_ops + dev->data->port_id, dev);
+
+       rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_NEW, NULL);
+
+       dev->state = RTE_ETH_DEV_ATTACHED;
+}
+
+int
+rte_eth_dev_release_port(struct rte_eth_dev *eth_dev)
+{
+       if (eth_dev == NULL)
+               return -EINVAL;
+
+       eth_dev_shared_data_prepare();
+
+       if (eth_dev->state != RTE_ETH_DEV_UNUSED)
+               rte_eth_dev_callback_process(eth_dev,
+                               RTE_ETH_EVENT_DESTROY, NULL);
+
+       eth_dev_fp_ops_reset(rte_eth_fp_ops + eth_dev->data->port_id);
+
+       rte_spinlock_lock(&eth_dev_shared_data->ownership_lock);
+
+       eth_dev->state = RTE_ETH_DEV_UNUSED;
+       eth_dev->device = NULL;
+       eth_dev->process_private = NULL;
+       eth_dev->intr_handle = NULL;
+       eth_dev->rx_pkt_burst = NULL;
+       eth_dev->tx_pkt_burst = NULL;
+       eth_dev->tx_pkt_prepare = NULL;
+       eth_dev->rx_queue_count = NULL;
+       eth_dev->rx_descriptor_status = NULL;
+       eth_dev->tx_descriptor_status = NULL;
+       eth_dev->dev_ops = NULL;
+
+       if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
+               rte_free(eth_dev->data->rx_queues);
+               rte_free(eth_dev->data->tx_queues);
+               rte_free(eth_dev->data->mac_addrs);
+               rte_free(eth_dev->data->hash_mac_addrs);
+               rte_free(eth_dev->data->dev_private);
+               pthread_mutex_destroy(&eth_dev->data->flow_ops_mutex);
+               memset(eth_dev->data, 0, sizeof(struct rte_eth_dev_data));
+       }
+
+       rte_spinlock_unlock(&eth_dev_shared_data->ownership_lock);
+
+       return 0;
+}
+
+int
+rte_eth_dev_create(struct rte_device *device, const char *name,
+       size_t priv_data_size,
+       ethdev_bus_specific_init ethdev_bus_specific_init,
+       void *bus_init_params,
+       ethdev_init_t ethdev_init, void *init_params)
+{
+       struct rte_eth_dev *ethdev;
+       int retval;
+
+       RTE_FUNC_PTR_OR_ERR_RET(*ethdev_init, -EINVAL);
+
+       if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
+               ethdev = rte_eth_dev_allocate(name);
+               if (!ethdev)
+                       return -ENODEV;
+
+               if (priv_data_size) {
+                       ethdev->data->dev_private = rte_zmalloc_socket(
+                               name, priv_data_size, RTE_CACHE_LINE_SIZE,
+                               device->numa_node);
+
+                       if (!ethdev->data->dev_private) {
+                               RTE_ETHDEV_LOG(ERR,
+                                       "failed to allocate private data\n");
+                               retval = -ENOMEM;
+                               goto probe_failed;
+                       }
+               }
+       } else {
+               ethdev = rte_eth_dev_attach_secondary(name);
+               if (!ethdev) {
+                       RTE_ETHDEV_LOG(ERR,
+                               "secondary process attach failed, ethdev doesn't exist\n");
+                       return  -ENODEV;
+               }
+       }
+
+       ethdev->device = device;
+
+       if (ethdev_bus_specific_init) {
+               retval = ethdev_bus_specific_init(ethdev, bus_init_params);
+               if (retval) {
+                       RTE_ETHDEV_LOG(ERR,
+                               "ethdev bus specific initialisation failed\n");
+                       goto probe_failed;
+               }
+       }
+
+       retval = ethdev_init(ethdev, init_params);
+       if (retval) {
+               RTE_ETHDEV_LOG(ERR, "ethdev initialisation failed\n");
+               goto probe_failed;
+       }
+
+       rte_eth_dev_probing_finish(ethdev);
+
+       return retval;
+
+probe_failed:
+       rte_eth_dev_release_port(ethdev);
+       return retval;
+}
+
+int
+rte_eth_dev_destroy(struct rte_eth_dev *ethdev,
+       ethdev_uninit_t ethdev_uninit)
+{
+       int ret;
+
+       ethdev = rte_eth_dev_allocated(ethdev->data->name);
+       if (!ethdev)
+               return -ENODEV;
+
+       RTE_FUNC_PTR_OR_ERR_RET(*ethdev_uninit, -EINVAL);
+
+       ret = ethdev_uninit(ethdev);
+       if (ret)
+               return ret;
+
+       return rte_eth_dev_release_port(ethdev);
+}
+
+struct rte_eth_dev *
+rte_eth_dev_get_by_name(const char *name)
+{
+       uint16_t pid;
+
+       if (rte_eth_dev_get_port_by_name(name, &pid))
+               return NULL;
+
+       return &rte_eth_devices[pid];
+}
+
+int
+rte_eth_dev_is_rx_hairpin_queue(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+       if (dev->data->rx_queue_state[queue_id] == RTE_ETH_QUEUE_STATE_HAIRPIN)
+               return 1;
+       return 0;
+}
+
+int
+rte_eth_dev_is_tx_hairpin_queue(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+       if (dev->data->tx_queue_state[queue_id] == RTE_ETH_QUEUE_STATE_HAIRPIN)
+               return 1;
+       return 0;
+}
+
+void
+rte_eth_dev_internal_reset(struct rte_eth_dev *dev)
+{
+       if (dev->data->dev_started) {
+               RTE_ETHDEV_LOG(ERR, "Port %u must be stopped to allow reset\n",
+                       dev->data->port_id);
+               return;
+       }
+
+       eth_dev_rx_queue_config(dev, 0);
+       eth_dev_tx_queue_config(dev, 0);
+
+       memset(&dev->data->dev_conf, 0, sizeof(dev->data->dev_conf));
+}
+
+static int
+eth_dev_devargs_tokenise(struct rte_kvargs *arglist, const char *str_in)
+{
+       int state;
+       struct rte_kvargs_pair *pair;
+       char *letter;
+
+       arglist->str = strdup(str_in);
+       if (arglist->str == NULL)
+               return -ENOMEM;
+
+       letter = arglist->str;
+       state = 0;
+       arglist->count = 0;
+       pair = &arglist->pairs[0];
+       while (1) {
+               switch (state) {
+               case 0: /* Initial */
+                       if (*letter == '=')
+                               return -EINVAL;
+                       else if (*letter == '\0')
+                               return 0;
+
+                       state = 1;
+                       pair->key = letter;
+                       /* fallthrough */
+
+               case 1: /* Parsing key */
+                       if (*letter == '=') {
+                               *letter = '\0';
+                               pair->value = letter + 1;
+                               state = 2;
+                       } else if (*letter == ',' || *letter == '\0')
+                               return -EINVAL;
+                       break;
+
+
+               case 2: /* Parsing value */
+                       if (*letter == '[')
+                               state = 3;
+                       else if (*letter == ',') {
+                               *letter = '\0';
+                               arglist->count++;
+                               pair = &arglist->pairs[arglist->count];
+                               state = 0;
+                       } else if (*letter == '\0') {
+                               letter--;
+                               arglist->count++;
+                               pair = &arglist->pairs[arglist->count];
+                               state = 0;
+                       }
+                       break;
+
+               case 3: /* Parsing list */
+                       if (*letter == ']')
+                               state = 2;
+                       else if (*letter == '\0')
+                               return -EINVAL;
+                       break;
+               }
+               letter++;
+       }
+}
+
+int
+rte_eth_devargs_parse(const char *dargs, struct rte_eth_devargs *eth_da)
+{
+       struct rte_kvargs args;
+       struct rte_kvargs_pair *pair;
+       unsigned int i;
+       int result = 0;
+
+       memset(eth_da, 0, sizeof(*eth_da));
+
+       result = eth_dev_devargs_tokenise(&args, dargs);
+       if (result < 0)
+               goto parse_cleanup;
+
+       for (i = 0; i < args.count; i++) {
+               pair = &args.pairs[i];
+               if (strcmp("representor", pair->key) == 0) {
+                       if (eth_da->type != RTE_ETH_REPRESENTOR_NONE) {
+                               RTE_LOG(ERR, EAL, "duplicated representor key: %s\n",
+                                       dargs);
+                               result = -1;
+                               goto parse_cleanup;
+                       }
+                       result = rte_eth_devargs_parse_representor_ports(
+                                       pair->value, eth_da);
+                       if (result < 0)
+                               goto parse_cleanup;
+               }
+       }
+
+parse_cleanup:
+       free(args.str);
+
+       return result;
+}
+
+static inline int
+eth_dev_dma_mzone_name(char *name, size_t len, uint16_t port_id, uint16_t queue_id,
+               const char *ring_name)
+{
+       return snprintf(name, len, "eth_p%d_q%d_%s",
+                       port_id, queue_id, ring_name);
+}
+
+int
+rte_eth_dma_zone_free(const struct rte_eth_dev *dev, const char *ring_name,
+               uint16_t queue_id)
+{
+       char z_name[RTE_MEMZONE_NAMESIZE];
+       const struct rte_memzone *mz;
+       int rc = 0;
+
+       rc = eth_dev_dma_mzone_name(z_name, sizeof(z_name), dev->data->port_id,
+                       queue_id, ring_name);
+       if (rc >= RTE_MEMZONE_NAMESIZE) {
+               RTE_ETHDEV_LOG(ERR, "ring name too long\n");
+               return -ENAMETOOLONG;
+       }
+
+       mz = rte_memzone_lookup(z_name);
+       if (mz)
+               rc = rte_memzone_free(mz);
+       else
+               rc = -ENOENT;
+
+       return rc;
+}
+
+const struct rte_memzone *
+rte_eth_dma_zone_reserve(const struct rte_eth_dev *dev, const char *ring_name,
+                        uint16_t queue_id, size_t size, unsigned int align,
+                        int socket_id)
+{
+       char z_name[RTE_MEMZONE_NAMESIZE];
+       const struct rte_memzone *mz;
+       int rc;
+
+       rc = eth_dev_dma_mzone_name(z_name, sizeof(z_name), dev->data->port_id,
+                       queue_id, ring_name);
+       if (rc >= RTE_MEMZONE_NAMESIZE) {
+               RTE_ETHDEV_LOG(ERR, "ring name too long\n");
+               rte_errno = ENAMETOOLONG;
+               return NULL;
+       }
+
+       mz = rte_memzone_lookup(z_name);
+       if (mz) {
+               if ((socket_id != SOCKET_ID_ANY && socket_id != mz->socket_id) ||
+                               size > mz->len ||
+                               ((uintptr_t)mz->addr & (align - 1)) != 0) {
+                       RTE_ETHDEV_LOG(ERR,
+                               "memzone %s does not justify the requested attributes\n",
+                               mz->name);
+                       return NULL;
+               }
+
+               return mz;
+       }
+
+       return rte_memzone_reserve_aligned(z_name, size, socket_id,
+                       RTE_MEMZONE_IOVA_CONTIG, align);
+}
+
+int
+rte_eth_hairpin_queue_peer_bind(uint16_t cur_port, uint16_t cur_queue,
+                               struct rte_hairpin_peer_info *peer_info,
+                               uint32_t direction)
+{
+       struct rte_eth_dev *dev;
+
+       if (peer_info == NULL)
+               return -EINVAL;
+
+       /* No need to check the validity again. */
+       dev = &rte_eth_devices[cur_port];
+       RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->hairpin_queue_peer_bind,
+                               -ENOTSUP);
+
+       return (*dev->dev_ops->hairpin_queue_peer_bind)(dev, cur_queue,
+                                                       peer_info, direction);
+}
+
+int
+rte_eth_hairpin_queue_peer_unbind(uint16_t cur_port, uint16_t cur_queue,
+                                 uint32_t direction)
+{
+       struct rte_eth_dev *dev;
+
+       /* No need to check the validity again. */
+       dev = &rte_eth_devices[cur_port];
+       RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->hairpin_queue_peer_unbind,
+                               -ENOTSUP);
+
+       return (*dev->dev_ops->hairpin_queue_peer_unbind)(dev, cur_queue,
+                                                         direction);
+}
+
+int
+rte_eth_hairpin_queue_peer_update(uint16_t peer_port, uint16_t peer_queue,
+                                 struct rte_hairpin_peer_info *cur_info,
+                                 struct rte_hairpin_peer_info *peer_info,
+                                 uint32_t direction)
+{
+       struct rte_eth_dev *dev;
+
+       /* Current queue information is not mandatory. */
+       if (peer_info == NULL)
+               return -EINVAL;
+
+       /* No need to check the validity again. */
+       dev = &rte_eth_devices[peer_port];
+       RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->hairpin_queue_peer_update,
+                               -ENOTSUP);
+
+       return (*dev->dev_ops->hairpin_queue_peer_update)(dev, peer_queue,
+                                       cur_info, peer_info, direction);
+}
+
+int
+rte_eth_ip_reassembly_dynfield_register(int *field_offset, int *flag_offset)
+{
+       static const struct rte_mbuf_dynfield field_desc = {
+               .name = RTE_MBUF_DYNFIELD_IP_REASSEMBLY_NAME,
+               .size = sizeof(rte_eth_ip_reassembly_dynfield_t),
+               .align = __alignof__(rte_eth_ip_reassembly_dynfield_t),
+       };
+       static const struct rte_mbuf_dynflag ip_reassembly_dynflag = {
+               .name = RTE_MBUF_DYNFLAG_IP_REASSEMBLY_INCOMPLETE_NAME,
+       };
+       int offset;
+
+       offset = rte_mbuf_dynfield_register(&field_desc);
+       if (offset < 0)
+               return -1;
+       if (field_offset != NULL)
+               *field_offset = offset;
+
+       offset = rte_mbuf_dynflag_register(&ip_reassembly_dynflag);
+       if (offset < 0)
+               return -1;
+       if (flag_offset != NULL)
+               *flag_offset = offset;
+
+       return 0;
+}
 
 uint16_t
 rte_eth_pkt_burst_dummy(void *queue __rte_unused,
@@ -11,3 +636,134 @@ rte_eth_pkt_burst_dummy(void *queue __rte_unused,
 {
        return 0;
 }
+
+int
+rte_eth_representor_id_get(uint16_t port_id,
+                          enum rte_eth_representor_type type,
+                          int controller, int pf, int representor_port,
+                          uint16_t *repr_id)
+{
+       int ret, n, count;
+       uint32_t i;
+       struct rte_eth_representor_info *info = NULL;
+       size_t size;
+
+       if (type == RTE_ETH_REPRESENTOR_NONE)
+               return 0;
+       if (repr_id == NULL)
+               return -EINVAL;
+
+       /* Get PMD representor range info. */
+       ret = rte_eth_representor_info_get(port_id, NULL);
+       if (ret == -ENOTSUP && type == RTE_ETH_REPRESENTOR_VF &&
+           controller == -1 && pf == -1) {
+               /* Direct mapping for legacy VF representor. */
+               *repr_id = representor_port;
+               return 0;
+       } else if (ret < 0) {
+               return ret;
+       }
+       n = ret;
+       size = sizeof(*info) + n * sizeof(info->ranges[0]);
+       info = calloc(1, size);
+       if (info == NULL)
+               return -ENOMEM;
+       info->nb_ranges_alloc = n;
+       ret = rte_eth_representor_info_get(port_id, info);
+       if (ret < 0)
+               goto out;
+
+       /* Default controller and pf to caller. */
+       if (controller == -1)
+               controller = info->controller;
+       if (pf == -1)
+               pf = info->pf;
+
+       /* Locate representor ID. */
+       ret = -ENOENT;
+       for (i = 0; i < info->nb_ranges; ++i) {
+               if (info->ranges[i].type != type)
+                       continue;
+               if (info->ranges[i].controller != controller)
+                       continue;
+               if (info->ranges[i].id_end < info->ranges[i].id_base) {
+                       RTE_LOG(WARNING, EAL, "Port %hu invalid representor ID Range %u - %u, entry %d\n",
+                               port_id, info->ranges[i].id_base,
+                               info->ranges[i].id_end, i);
+                       continue;
+
+               }
+               count = info->ranges[i].id_end - info->ranges[i].id_base + 1;
+               switch (info->ranges[i].type) {
+               case RTE_ETH_REPRESENTOR_PF:
+                       if (pf < info->ranges[i].pf ||
+                           pf >= info->ranges[i].pf + count)
+                               continue;
+                       *repr_id = info->ranges[i].id_base +
+                                  (pf - info->ranges[i].pf);
+                       ret = 0;
+                       goto out;
+               case RTE_ETH_REPRESENTOR_VF:
+                       if (info->ranges[i].pf != pf)
+                               continue;
+                       if (representor_port < info->ranges[i].vf ||
+                           representor_port >= info->ranges[i].vf + count)
+                               continue;
+                       *repr_id = info->ranges[i].id_base +
+                                  (representor_port - info->ranges[i].vf);
+                       ret = 0;
+                       goto out;
+               case RTE_ETH_REPRESENTOR_SF:
+                       if (info->ranges[i].pf != pf)
+                               continue;
+                       if (representor_port < info->ranges[i].sf ||
+                           representor_port >= info->ranges[i].sf + count)
+                               continue;
+                       *repr_id = info->ranges[i].id_base +
+                             (representor_port - info->ranges[i].sf);
+                       ret = 0;
+                       goto out;
+               default:
+                       break;
+               }
+       }
+out:
+       free(info);
+       return ret;
+}
+
+int
+rte_eth_switch_domain_alloc(uint16_t *domain_id)
+{
+       uint16_t i;
+
+       *domain_id = RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID;
+
+       for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
+               if (eth_dev_switch_domains[i].state ==
+                       RTE_ETH_SWITCH_DOMAIN_UNUSED) {
+                       eth_dev_switch_domains[i].state =
+                               RTE_ETH_SWITCH_DOMAIN_ALLOCATED;
+                       *domain_id = i;
+                       return 0;
+               }
+       }
+
+       return -ENOSPC;
+}
+
+int
+rte_eth_switch_domain_free(uint16_t domain_id)
+{
+       if (domain_id == RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID ||
+               domain_id >= RTE_MAX_ETHPORTS)
+               return -EINVAL;
+
+       if (eth_dev_switch_domains[domain_id].state !=
+               RTE_ETH_SWITCH_DOMAIN_ALLOCATED)
+               return -EINVAL;
+
+       eth_dev_switch_domains[domain_id].state = RTE_ETH_SWITCH_DOMAIN_UNUSED;
+
+       return 0;
+}