Quelle  ice_eswitch.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0
/* Copyright (C) 2019-2021, Intel Corporation. */

#include "ice.h"
#include "ice_lib.h"
#include "ice_eswitch.h"
#include "ice_eswitch_br.h"
#include "ice_fltr.h"
#include "ice_repr.h"
#include "devlink/devlink.h"
#include "ice_tc_lib.h"

/**
 * ice_eswitch_setup_env - configure eswitch HW filters
 * @pf: pointer to PF struct
 *
 * This function adds HW filters configuration specific for switchdev
 * mode.
 */
static int ice_eswitch_setup_env(struct ice_pf *pf)
{
 struct ice_vsi *uplink_vsi = pf->eswitch.uplink_vsi;
 struct net_device *netdev = uplink_vsi->netdev;
 bool if_running = netif_running(netdev);
 struct ice_vsi_vlan_ops *vlan_ops;

 if (if_running && !test_and_set_bit(ICE_VSI_DOWN, uplink_vsi->state))
  if (ice_down(uplink_vsi))
   return -ENODEV;

 ice_remove_vsi_fltr(&pf->hw, uplink_vsi->idx);
 ice_vsi_cfg_sw_lldp(uplink_vsi, true, false);

 netif_addr_lock_bh(netdev);
 __dev_uc_unsync(netdev, NULL);
 __dev_mc_unsync(netdev, NULL);
 netif_addr_unlock_bh(netdev);

 if (ice_vsi_add_vlan_zero(uplink_vsi))
  goto err_vlan_zero;

 if (ice_set_dflt_vsi(uplink_vsi))
  goto err_def_rx;

 if (ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, true,
        ICE_FLTR_TX))
  goto err_def_tx;

 vlan_ops = ice_get_compat_vsi_vlan_ops(uplink_vsi);
 if (vlan_ops->dis_rx_filtering(uplink_vsi))
  goto err_vlan_filtering;

 if (ice_vsi_update_local_lb(uplink_vsi, true))
  goto err_override_local_lb;

 if (if_running && ice_up(uplink_vsi))
  goto err_up;

 return 0;

err_up:
 ice_vsi_update_local_lb(uplink_vsi, false);
err_override_local_lb:
 vlan_ops->ena_rx_filtering(uplink_vsi);
err_vlan_filtering:
 ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
    ICE_FLTR_TX);
err_def_tx:
 ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
    ICE_FLTR_RX);
err_def_rx:
 ice_vsi_del_vlan_zero(uplink_vsi);
err_vlan_zero:
 ice_fltr_add_mac_and_broadcast(uplink_vsi,
           uplink_vsi->port_info->mac.perm_addr,
           ICE_FWD_TO_VSI);
 if (if_running)
  ice_up(uplink_vsi);

 return -ENODEV;
}

/**
 * ice_eswitch_release_repr - clear PR VSI configuration
 * @pf: poiner to PF struct
 * @repr: pointer to PR
 */
static void
ice_eswitch_release_repr(struct ice_pf *pf, struct ice_repr *repr)
{
 struct ice_vsi *vsi = repr->src_vsi;

 /* Skip representors that aren't configured */
 if (!repr->dst)
  return;

 ice_vsi_update_security(vsi, ice_vsi_ctx_set_antispoof);
 metadata_dst_free(repr->dst);
 repr->dst = NULL;
 ice_fltr_add_mac_and_broadcast(vsi, repr->parent_mac,
           ICE_FWD_TO_VSI);
}

/**
 * ice_eswitch_setup_repr - configure PR to run in switchdev mode
 * @pf: pointer to PF struct
 * @repr: pointer to PR struct
 */
static int ice_eswitch_setup_repr(struct ice_pf *pf, struct ice_repr *repr)
{
 struct ice_vsi *uplink_vsi = pf->eswitch.uplink_vsi;
 struct ice_vsi *vsi = repr->src_vsi;
 struct metadata_dst *dst;

 repr->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX,
           GFP_KERNEL);
 if (!repr->dst)
  return -ENOMEM;

 netif_keep_dst(uplink_vsi->netdev);

 dst = repr->dst;
 dst->u.port_info.port_id = vsi->vsi_num;
 dst->u.port_info.lower_dev = uplink_vsi->netdev;

 return 0;
}

/**
 * ice_eswitch_cfg_vsi - configure VSI to work in slow-path
 * @vsi: VSI structure of representee
 * @mac: representee MAC
 *
 * Return: 0 on success, non-zero on error.
 */
int ice_eswitch_cfg_vsi(struct ice_vsi *vsi, const u8 *mac)
{
 int err;

 ice_remove_vsi_fltr(&vsi->back->hw, vsi->idx);

 err = ice_vsi_update_security(vsi, ice_vsi_ctx_clear_antispoof);
 if (err)
  goto err_update_security;

 err = ice_vsi_add_vlan_zero(vsi);
 if (err)
  goto err_vlan_zero;

 return 0;

err_vlan_zero:
 ice_vsi_update_security(vsi, ice_vsi_ctx_set_antispoof);
err_update_security:
 ice_fltr_add_mac_and_broadcast(vsi, mac, ICE_FWD_TO_VSI);

 return err;
}

/**
 * ice_eswitch_decfg_vsi - unroll changes done to VSI for switchdev
 * @vsi: VSI structure of representee
 * @mac: representee MAC
 */
void ice_eswitch_decfg_vsi(struct ice_vsi *vsi, const u8 *mac)
{
 ice_vsi_update_security(vsi, ice_vsi_ctx_set_antispoof);
 ice_fltr_add_mac_and_broadcast(vsi, mac, ICE_FWD_TO_VSI);
}

/**
 * ice_eswitch_update_repr - reconfigure port representor
 * @repr_id: representor ID
 * @vsi: VSI for which port representor is configured
 */
void ice_eswitch_update_repr(unsigned long *repr_id, struct ice_vsi *vsi)
{
 struct ice_pf *pf = vsi->back;
 struct ice_repr *repr;
 int err;

 if (!ice_is_switchdev_running(pf))
  return;

 repr = xa_load(&pf->eswitch.reprs, *repr_id);
 if (!repr)
  return;

 repr->src_vsi = vsi;
 repr->dst->u.port_info.port_id = vsi->vsi_num;

 if (repr->br_port)
  repr->br_port->vsi = vsi;

 err = ice_eswitch_cfg_vsi(vsi, repr->parent_mac);
 if (err)
  dev_err(ice_pf_to_dev(pf), "Failed to update VSI of port representor %d",
   repr->id);

 /* The VSI number is different, reload the PR with new id */
 if (repr->id != vsi->vsi_num) {
  xa_erase(&pf->eswitch.reprs, repr->id);
  repr->id = vsi->vsi_num;
  if (xa_insert(&pf->eswitch.reprs, repr->id, repr, GFP_KERNEL))
   dev_err(ice_pf_to_dev(pf), "Failed to reload port representor %d",
    repr->id);
  *repr_id = repr->id;
 }
}

/**
 * ice_eswitch_port_start_xmit - callback for packets transmit
 * @skb: send buffer
 * @netdev: network interface device structure
 *
 * Returns NETDEV_TX_OK if sent, else an error code
 */
netdev_tx_t
ice_eswitch_port_start_xmit(struct sk_buff *skb, struct net_device *netdev)
{
 struct ice_repr *repr = ice_netdev_to_repr(netdev);
 unsigned int len = skb->len;
 int ret;

 skb_dst_drop(skb);
 dst_hold((struct dst_entry *)repr->dst);
 skb_dst_set(skb, (struct dst_entry *)repr->dst);
 skb->dev = repr->dst->u.port_info.lower_dev;

 ret = dev_queue_xmit(skb);
 ice_repr_inc_tx_stats(repr, len, ret);

 return ret;
}

/**
 * ice_eswitch_set_target_vsi - set eswitch context in Tx context descriptor
 * @skb: pointer to send buffer
 * @off: pointer to offload struct
 */
void
ice_eswitch_set_target_vsi(struct sk_buff *skb,
      struct ice_tx_offload_params *off)
{
 struct metadata_dst *dst = skb_metadata_dst(skb);
 u64 cd_cmd, dst_vsi;

 if (!dst) {
  struct ethhdr *eth = (struct ethhdr *)skb_mac_header(skb);

  if (unlikely(eth->h_proto == htons(ETH_P_LLDP)))
   return;
  cd_cmd = ICE_TX_CTX_DESC_SWTCH_UPLINK << ICE_TXD_CTX_QW1_CMD_S;
  off->cd_qw1 |= (cd_cmd | ICE_TX_DESC_DTYPE_CTX);
 } else {
  cd_cmd = ICE_TX_CTX_DESC_SWTCH_VSI << ICE_TXD_CTX_QW1_CMD_S;
  dst_vsi = FIELD_PREP(ICE_TXD_CTX_QW1_VSI_M,
         dst->u.port_info.port_id);
  off->cd_qw1 = cd_cmd | dst_vsi | ICE_TX_DESC_DTYPE_CTX;
 }
}

/**
 * ice_eswitch_release_env - clear eswitch HW filters
 * @pf: pointer to PF struct
 *
 * This function removes HW filters configuration specific for switchdev
 * mode and restores default legacy mode settings.
 */
static void ice_eswitch_release_env(struct ice_pf *pf)
{
 struct ice_vsi *uplink_vsi = pf->eswitch.uplink_vsi;
 struct ice_vsi_vlan_ops *vlan_ops;

 vlan_ops = ice_get_compat_vsi_vlan_ops(uplink_vsi);

 ice_vsi_update_local_lb(uplink_vsi, false);
 vlan_ops->ena_rx_filtering(uplink_vsi);
 ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
    ICE_FLTR_TX);
 ice_cfg_dflt_vsi(uplink_vsi->port_info, uplink_vsi->idx, false,
    ICE_FLTR_RX);
 ice_fltr_add_mac_and_broadcast(uplink_vsi,
           uplink_vsi->port_info->mac.perm_addr,
           ICE_FWD_TO_VSI);
 ice_vsi_cfg_sw_lldp(uplink_vsi, true, true);
}

/**
 * ice_eswitch_enable_switchdev - configure eswitch in switchdev mode
 * @pf: pointer to PF structure
 */
static int ice_eswitch_enable_switchdev(struct ice_pf *pf)
{
 struct ice_vsi *uplink_vsi;

 uplink_vsi = ice_get_main_vsi(pf);
 if (!uplink_vsi)
  return -ENODEV;

 if (netif_is_any_bridge_port(uplink_vsi->netdev)) {
  dev_err(ice_pf_to_dev(pf),
   "Uplink port cannot be a bridge port\n");
  return -EINVAL;
 }

 pf->eswitch.uplink_vsi = uplink_vsi;

 if (ice_eswitch_setup_env(pf))
  return -ENODEV;

 if (ice_eswitch_br_offloads_init(pf))
  goto err_br_offloads;

 pf->eswitch.is_running = true;

 return 0;

err_br_offloads:
 ice_eswitch_release_env(pf);
 return -ENODEV;
}

/**
 * ice_eswitch_disable_switchdev - disable eswitch resources
 * @pf: pointer to PF structure
 */
static void ice_eswitch_disable_switchdev(struct ice_pf *pf)
{
 ice_eswitch_br_offloads_deinit(pf);
 ice_eswitch_release_env(pf);

 pf->eswitch.is_running = false;
}

/**
 * ice_eswitch_mode_set - set new eswitch mode
 * @devlink: pointer to devlink structure
 * @mode: eswitch mode to switch to
 * @extack: pointer to extack structure
 */
int
ice_eswitch_mode_set(struct devlink *devlink, u16 mode,
       struct netlink_ext_ack *extack)
{
 struct ice_pf *pf = devlink_priv(devlink);

 if (pf->eswitch_mode == mode)
  return 0;

 if (ice_has_vfs(pf)) {
  dev_info(ice_pf_to_dev(pf), "Changing eswitch mode is allowed only if there is no VFs created");
  NL_SET_ERR_MSG_MOD(extack, "Changing eswitch mode is allowed only if there is no VFs created");
  return -EOPNOTSUPP;
 }

 switch (mode) {
 case DEVLINK_ESWITCH_MODE_LEGACY:
  dev_info(ice_pf_to_dev(pf), "PF %d changed eswitch mode to legacy",
    pf->hw.pf_id);
  xa_destroy(&pf->eswitch.reprs);
  NL_SET_ERR_MSG_MOD(extack, "Changed eswitch mode to legacy");
  break;
 case DEVLINK_ESWITCH_MODE_SWITCHDEV:
 {
  if (ice_is_adq_active(pf)) {
   dev_err(ice_pf_to_dev(pf), "Couldn't change eswitch mode to switchdev - ADQ is active. Delete ADQ configs and try again, e.g. tc qdisc del dev $PF root");
   NL_SET_ERR_MSG_MOD(extack, "Couldn't change eswitch mode to switchdev - ADQ is active. Delete ADQ configs and try again, e.g. tc qdisc del dev $PF root");
   return -EOPNOTSUPP;
  }

  dev_info(ice_pf_to_dev(pf), "PF %d changed eswitch mode to switchdev",
    pf->hw.pf_id);
  xa_init(&pf->eswitch.reprs);
  NL_SET_ERR_MSG_MOD(extack, "Changed eswitch mode to switchdev");
  break;
 }
 default:
  NL_SET_ERR_MSG_MOD(extack, "Unknown eswitch mode");
  return -EINVAL;
 }

 pf->eswitch_mode = mode;
 return 0;
}

/**
 * ice_eswitch_mode_get - get current eswitch mode
 * @devlink: pointer to devlink structure
 * @mode: output parameter for current eswitch mode
 */
int ice_eswitch_mode_get(struct devlink *devlink, u16 *mode)
{
 struct ice_pf *pf = devlink_priv(devlink);

 *mode = pf->eswitch_mode;
 return 0;
}

/**
 * ice_is_eswitch_mode_switchdev - check if eswitch mode is set to switchdev
 * @pf: pointer to PF structure
 *
 * Returns true if eswitch mode is set to DEVLINK_ESWITCH_MODE_SWITCHDEV,
 * false otherwise.
 */
bool ice_is_eswitch_mode_switchdev(struct ice_pf *pf)
{
 return pf->eswitch_mode == DEVLINK_ESWITCH_MODE_SWITCHDEV;
}

/**
 * ice_eswitch_start_all_tx_queues - start Tx queues of all port representors
 * @pf: pointer to PF structure
 */
static void ice_eswitch_start_all_tx_queues(struct ice_pf *pf)
{
 struct ice_repr *repr;
 unsigned long id;

 if (test_bit(ICE_DOWN, pf->state))
  return;

 xa_for_each(&pf->eswitch.reprs, id, repr)
  ice_repr_start_tx_queues(repr);
}

/**
 * ice_eswitch_stop_all_tx_queues - stop Tx queues of all port representors
 * @pf: pointer to PF structure
 */
void ice_eswitch_stop_all_tx_queues(struct ice_pf *pf)
{
 struct ice_repr *repr;
 unsigned long id;

 if (test_bit(ICE_DOWN, pf->state))
  return;

 xa_for_each(&pf->eswitch.reprs, id, repr)
  ice_repr_stop_tx_queues(repr);
}

static void ice_eswitch_stop_reprs(struct ice_pf *pf)
{
 ice_eswitch_stop_all_tx_queues(pf);
}

static void ice_eswitch_start_reprs(struct ice_pf *pf)
{
 ice_eswitch_start_all_tx_queues(pf);
}

static int
ice_eswitch_attach(struct ice_pf *pf, struct ice_repr *repr, unsigned long *id)
{
 int err;

 if (pf->eswitch_mode == DEVLINK_ESWITCH_MODE_LEGACY)
  return 0;

 if (xa_empty(&pf->eswitch.reprs)) {
  err = ice_eswitch_enable_switchdev(pf);
  if (err)
   return err;
 }

 ice_eswitch_stop_reprs(pf);

 err = repr->ops.add(repr);
 if (err)
  goto err_create_repr;

 err = ice_eswitch_setup_repr(pf, repr);
 if (err)
  goto err_setup_repr;

 err = xa_insert(&pf->eswitch.reprs, repr->id, repr, GFP_KERNEL);
 if (err)
  goto err_xa_alloc;

 *id = repr->id;

 ice_eswitch_start_reprs(pf);

 return 0;

err_xa_alloc:
 ice_eswitch_release_repr(pf, repr);
err_setup_repr:
 repr->ops.rem(repr);
err_create_repr:
 if (xa_empty(&pf->eswitch.reprs))
  ice_eswitch_disable_switchdev(pf);
 ice_eswitch_start_reprs(pf);

 return err;
}

/**
 * ice_eswitch_attach_vf - attach VF to a eswitch
 * @pf: pointer to PF structure
 * @vf: pointer to VF structure to be attached
 *
 * During attaching port representor for VF is created.
 *
 * Return: zero on success or an error code on failure.
 */
int ice_eswitch_attach_vf(struct ice_pf *pf, struct ice_vf *vf)
{
 struct devlink *devlink = priv_to_devlink(pf);
 struct ice_repr *repr;
 int err;

 if (!ice_is_eswitch_mode_switchdev(pf))
  return 0;

 repr = ice_repr_create_vf(vf);
 if (IS_ERR(repr))
  return PTR_ERR(repr);

 devl_lock(devlink);
 err = ice_eswitch_attach(pf, repr, &vf->repr_id);
 if (err)
  ice_repr_destroy(repr);
 devl_unlock(devlink);

 return err;
}

/**
 * ice_eswitch_attach_sf - attach SF to a eswitch
 * @pf: pointer to PF structure
 * @sf: pointer to SF structure to be attached
 *
 * During attaching port representor for SF is created.
 *
 * Return: zero on success or an error code on failure.
 */
int ice_eswitch_attach_sf(struct ice_pf *pf, struct ice_dynamic_port *sf)
{
 struct ice_repr *repr = ice_repr_create_sf(sf);
 int err;

 if (IS_ERR(repr))
  return PTR_ERR(repr);

 err = ice_eswitch_attach(pf, repr, &sf->repr_id);
 if (err)
  ice_repr_destroy(repr);

 return err;
}

static void ice_eswitch_detach(struct ice_pf *pf, struct ice_repr *repr)
{
 ice_eswitch_stop_reprs(pf);
 repr->ops.rem(repr);

 xa_erase(&pf->eswitch.reprs, repr->id);

 if (xa_empty(&pf->eswitch.reprs))
  ice_eswitch_disable_switchdev(pf);

 ice_eswitch_release_repr(pf, repr);
 ice_repr_destroy(repr);

 if (xa_empty(&pf->eswitch.reprs)) {
  struct devlink *devlink = priv_to_devlink(pf);

  /* since all port representors are destroyed, there is
 * no point in keeping the nodes
 */
  ice_devlink_rate_clear_tx_topology(ice_get_main_vsi(pf));
  devl_rate_nodes_destroy(devlink);
 } else {
  ice_eswitch_start_reprs(pf);
 }
}

/**
 * ice_eswitch_detach_vf - detach VF from a eswitch
 * @pf: pointer to PF structure
 * @vf: pointer to VF structure to be detached
 */
void ice_eswitch_detach_vf(struct ice_pf *pf, struct ice_vf *vf)
{
 struct ice_repr *repr = xa_load(&pf->eswitch.reprs, vf->repr_id);
 struct devlink *devlink = priv_to_devlink(pf);

 if (!repr)
  return;

 devl_lock(devlink);
 ice_eswitch_detach(pf, repr);
 devl_unlock(devlink);
}

/**
 * ice_eswitch_detach_sf - detach SF from a eswitch
 * @pf: pointer to PF structure
 * @sf: pointer to SF structure to be detached
 */
void ice_eswitch_detach_sf(struct ice_pf *pf, struct ice_dynamic_port *sf)
{
 struct ice_repr *repr = xa_load(&pf->eswitch.reprs, sf->repr_id);

 if (!repr)
  return;

 ice_eswitch_detach(pf, repr);
}

/**
 * ice_eswitch_get_target - get netdev based on src_vsi from descriptor
 * @rx_ring: ring used to receive the packet
 * @rx_desc: descriptor used to get src_vsi value
 *
 * Get src_vsi value from descriptor and load correct representor. If it isn't
 * found return rx_ring->netdev.
 */
struct net_device *ice_eswitch_get_target(struct ice_rx_ring *rx_ring,
       union ice_32b_rx_flex_desc *rx_desc)
{
 struct ice_eswitch *eswitch = &rx_ring->vsi->back->eswitch;
 struct ice_32b_rx_flex_desc_nic_2 *desc;
 struct ice_repr *repr;

 desc = (struct ice_32b_rx_flex_desc_nic_2 *)rx_desc;
 repr = xa_load(&eswitch->reprs, le16_to_cpu(desc->src_vsi));
 if (!repr)
  return rx_ring->netdev;

 return repr->netdev;
}