| Message ID | 20181218215456.15494-1-harshitha.ramamurthy@intel.com |
|---|---|
| State | Changes Requested |
| Delegated to: | Jeff Kirsher |
| Headers | show |
| Series | i40e: Add macvlan support on i40e | expand |
> -----Original Message----- > From: Intel-wired-lan [mailto:intel-wired-lan-bounces@osuosl.org] On > Behalf Of Harshitha Ramamurthy > Sent: Tuesday, December 18, 2018 1:55 PM > To: intel-wired-lan@lists.osuosl.org > Cc: Duyck, Alexander H <alexander.h.duyck@intel.com> > Subject: [Intel-wired-lan] [PATCH] i40e: Add macvlan support on i40e > > This patch enables macvlan offloads on Fortville devices. The idea is to use > channels as macvlan interfaces. The channels are VSIs of type VMDQ. When > the first macvlan is created, the maximum no. of channels possible are > created. From then on, as a macvlan interface is created, a mac filter is added > to these already created channels (VSIs). > > This patch builds on top of the recent changes which move away from the > select_queue implementation of picking the tx queue. > > Steps to configure the macvlan: > 1. sudo ethtool -K ens261f1 l2-fwd-offload on 2. ip link add link ens261f1 > name macvlan1 type macvlan 3. sudo ip link add link ens261f1 name macvlan1 > type macvlan 4. sudo ip link set macvlan1 up > > Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com> > --- > drivers/net/ethernet/intel/i40e/i40e.h | 26 ++ > drivers/net/ethernet/intel/i40e/i40e_main.c | 425 +++++++++++++++++++- > 2 files changed, 449 insertions(+), 2 deletions(-) Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
On Tue, Dec 18, 2018 at 1:55 PM Harshitha Ramamurthy <harshitha.ramamurthy@intel.com> wrote: > > This patch enables macvlan offloads on Fortville devices. The idea > is to use channels as macvlan interfaces. The channels are VSIs of > type VMDQ. When the first macvlan is created, the maximum no. of > channels possible are created. From then on, as a macvlan interface > is created, a mac filter is added to these already created channels > (VSIs). > > This patch builds on top of the recent changes which move > away from the select_queue implementation of picking the tx queue. > > Steps to configure the macvlan: > 1. sudo ethtool -K ens261f1 l2-fwd-offload on > 2. ip link add link ens261f1 name macvlan1 type macvlan > 3. sudo ip link add link ens261f1 name macvlan1 type macvlan > 4. sudo ip link set macvlan1 up Nice to see this finally coming out, thanks. A few nit-pics below, otherwise Acked-by: Shannon Nelson <shannon.lee.nelson@gmail.com> Cheers, sln > > Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com> > --- > drivers/net/ethernet/intel/i40e/i40e.h | 26 ++ > drivers/net/ethernet/intel/i40e/i40e_main.c | 425 +++++++++++++++++++- > 2 files changed, 449 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h > index 4f4de85887a6..7416625e091e 100644 > --- a/drivers/net/ethernet/intel/i40e/i40e.h > +++ b/drivers/net/ethernet/intel/i40e/i40e.h > @@ -27,6 +27,7 @@ > #include <net/ip6_checksum.h> > #include <linux/ethtool.h> > #include <linux/if_vlan.h> > +#include <linux/if_macvlan.h> > #include <linux/if_bridge.h> > #include <linux/clocksource.h> > #include <linux/net_tstamp.h> > @@ -390,6 +391,11 @@ struct i40e_flex_pit { > u8 pit_index; > }; > > +struct i40e_fwd_adapter { > + struct net_device *netdev; > + int bit_no; > +}; > + > struct i40e_channel { > struct list_head list; > bool initialized; > @@ -404,11 +410,25 @@ struct i40e_channel { > struct i40e_aqc_vsi_properties_data info; > > u64 max_tx_rate; > + struct i40e_fwd_adapter *fwd; > > /* track this channel belongs to which VSI */ > struct i40e_vsi *parent_vsi; > }; > > +static inline bool i40e_is_channel_macvlan(struct i40e_channel *ch) > +{ > + return !!ch->fwd; > +} > + > +static inline u8 *i40e_channel_mac(struct i40e_channel *ch) > +{ > + if (i40e_is_channel_macvlan(ch)) > + return ch->fwd->netdev->dev_addr; > + else > + return NULL; > +} > + > /* struct that defines the Ethernet device */ > struct i40e_pf { > struct pci_dev *pdev; > @@ -784,6 +804,12 @@ struct i40e_vsi { > struct list_head ch_list; > u16 tc_seid_map[I40E_MAX_TRAFFIC_CLASS]; > > + /* macvlan fields */ > +#define I40E_MAX_MACVLANS 128 /* Max HW capable vectors - 1 on FVL */ > + DECLARE_BITMAP(fwd_bitmask, I40E_MAX_MACVLANS); > + struct list_head macvlan_list; > + int macvlan_cnt; > + > void *priv; /* client driver data reference. */ > > /* VSI specific handlers */ > diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c > index 1ab1f579343f..1e61c22c27f4 100644 > --- a/drivers/net/ethernet/intel/i40e/i40e_main.c > +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c > @@ -5818,8 +5818,10 @@ static int i40e_add_channel(struct i40e_pf *pf, u16 uplink_seid, > return -ENOENT; > } > > - /* Success, update channel */ > - ch->enabled_tc = enabled_tc; > + /* Success, update channel, set enabled_tc only if the channel > + * is not a macvlan > + */ > + ch->enabled_tc = !i40e_is_channel_macvlan(ch) && enabled_tc; > ch->seid = ctxt.seid; > ch->vsi_number = ctxt.vsi_number; > ch->stat_counter_idx = cpu_to_le16(ctxt.info.stat_counter_idx); > @@ -6811,6 +6813,417 @@ static void i40e_vsi_set_default_tc_config(struct i40e_vsi *vsi) > } > } > > +/** > + * i40e_del_macvlan_filter > + * @hw: pointer to the HW structure > + * @seid: seid of the channel VSI > + * @macaddr: the mac address to apply as a filter > + * @aq_err: store the admin Q error > + * > + * This function deletes a mac filter on the channel VSI which serves as the > + * macvlan. Returns 0 on success. > + **/ > +static i40e_status i40e_del_macvlan_filter(struct i40e_hw *hw, u16 seid, > + const u8 *macaddr, int *aq_err) > +{ > + struct i40e_aqc_remove_macvlan_element_data element; > + i40e_status status; > + > + memset(&element, 0, sizeof(element)); > + ether_addr_copy(element.mac_addr, macaddr); > + element.vlan_tag = 0; > + element.flags = I40E_AQC_MACVLAN_DEL_PERFECT_MATCH; > + status = i40e_aq_remove_macvlan(hw, seid, &element, 1, NULL); > + *aq_err = hw->aq.asq_last_status; > + return status; > +} > + > +/** > + * i40e_add_macvlan_filter > + * @hw: pointer to the HW structure > + * @seid: seid of the channel VSI > + * @macaddr: the mac address to apply as a filter > + * @aq_err: store the admin Q error > + * > + * This function adds a mac filter on the channel VSI which serves as the > + * macvlan. Returns 0 on success. > + **/ > +static i40e_status i40e_add_macvlan_filter(struct i40e_hw *hw, u16 seid, > + const u8 *macaddr, int *aq_err) > +{ > + struct i40e_aqc_add_macvlan_element_data element; > + i40e_status status; > + u16 cmd_flags = 0; > + > + ether_addr_copy(element.mac_addr, macaddr); > + element.vlan_tag = 0; > + element.queue_number = 0; > + element.match_method = I40E_AQC_MM_ERR_NO_RES; > + cmd_flags |= I40E_AQC_MACVLAN_ADD_PERFECT_MATCH; > + element.flags = cpu_to_le16(cmd_flags); > + status = i40e_aq_add_macvlan(hw, seid, &element, 1, NULL); > + *aq_err = hw->aq.asq_last_status; > + return status; > +} > + > +/** > + * i40e_fwd_ring_up - bring the macvlan device up > + * @vsi: the VSI we want to access > + * @vdev: macvlan netdevice > + * @fwd: the private fwd structure > + */ > +static int i40e_fwd_ring_up(struct i40e_vsi *vsi, struct net_device *vdev, > + struct i40e_fwd_adapter *fwd) > +{ > + int ret = 0, num_tc = 1, i, aq_err; > + struct i40e_channel *ch, *ch_tmp; > + struct i40e_pf *pf = vsi->back; > + struct i40e_hw *hw = &pf->hw; > + > + if (list_empty(&vsi->macvlan_list)) > + return -EINVAL; > + > + /* Go through the list and find an avaialble channel */ s/avaialble/available/ > + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, list) { > + if (!i40e_is_channel_macvlan(ch)) { Will the channel ever be used for anything else? Perhaps the sense of this should be to check for in use rather than is it a macvlan? > + ch->fwd = fwd; > + /* record configuration for macvlan interface in vdev */ > + for (i = 0; i < num_tc; i++) > + netdev_bind_sb_channel_queue(vsi->netdev, vdev, > + i, > + ch->num_queue_pairs, > + ch->base_queue); > + for (i = 0; i < ch->num_queue_pairs; i++) { > + struct i40e_ring *tx_ring, *rx_ring; > + u16 pf_q; > + > + pf_q = ch->base_queue + i; > + > + /* Get to TX ring ptr */ > + tx_ring = vsi->tx_rings[pf_q]; > + tx_ring->ch = ch; > + > + /* Get the RX ring ptr */ > + rx_ring = vsi->rx_rings[pf_q]; > + rx_ring->ch = ch; > + } > + break; > + } > + } > + > + /* Guarantee all rings are updated before we update the > + * MAC address filter. > + */ > + wmb(); > + > + /* Add a mac filter */ > + ret = i40e_add_macvlan_filter(hw, ch->seid, vdev->dev_addr, &aq_err); > + if (ret) { > + /* if we cannot add the MAC rule then disable the offload */ > + macvlan_release_l2fw_offload(vdev); > + for (i = 0; i < ch->num_queue_pairs; i++) { > + struct i40e_ring *rx_ring; > + u16 pf_q; > + > + pf_q = ch->base_queue + i; > + rx_ring = vsi->rx_rings[pf_q]; > + rx_ring->netdev = NULL; > + } > + dev_info(&pf->pdev->dev, > + "Error adding mac filter on macvlan err %s, aq_err %s\n", > + i40e_stat_str(hw, ret), > + i40e_aq_str(hw, aq_err)); > + netdev_err(vdev, "L2fwd offload disabled to L2 filter error\n"); > + } > + return ret; > +} > + > +/** > + * i40e_setup_macvlans - create the channels which will be macvlans > + * @vsi: the VSI we want to access > + * @macvlan_cnt: no. of macvlans to be setup > + * @qcnt: no. of Qs per macvlan > + * @vdev: macvlan netdevice > + */ > +static int i40e_setup_macvlans(struct i40e_vsi *vsi, u16 macvlan_cnt, u16 qcnt, > + struct net_device *vdev) > +{ > + struct i40e_pf *pf = vsi->back; > + struct i40e_hw *hw = &pf->hw; > + struct i40e_vsi_context ctxt; > + u16 sections, qmap, num_qps; > + struct i40e_channel *ch; > + int i, pow, ret = 0; > + u8 offset = 0; > + > + if (vsi->type != I40E_VSI_MAIN) > + return -EINVAL; > + if (!macvlan_cnt) > + return -EBUSY; EBUSY doesn't look right here, maybe EINVAL? > + > + num_qps = vsi->num_queue_pairs - (macvlan_cnt * qcnt); > + > + /* find the next higher power-of-2 of num queue pairs */ > + pow = fls(roundup_pow_of_two(num_qps) - 1); > + > + qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) | > + (pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT); > + > + /* Setup context bits for the main VSI */ > + sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID; > + sections |= I40E_AQ_VSI_PROP_SCHED_VALID; > + ctxt.seid = vsi->seid; Perhaps memset( 0 ) the ctxt before using it? > + ctxt.pf_num = vsi->back->hw.pf_id; > + ctxt.vf_num = 0; > + ctxt.uplink_seid = vsi->uplink_seid; > + ctxt.info = vsi->info; > + ctxt.info.tc_mapping[0] = cpu_to_le16(qmap); > + ctxt.info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG); > + ctxt.info.queue_mapping[0] = cpu_to_le16(vsi->base_queue); > + ctxt.info.valid_sections |= cpu_to_le16(sections); > + > + /* Reconfigure RSS for main VSI with max queue count */ s/with max/with new max/ > + vsi->rss_size = max_t(u16, num_qps, qcnt); > + ret = i40e_vsi_config_rss(vsi); > + if (ret) { > + dev_info(&vsi->back->pdev->dev, > + "Failed to reconfig rss for num_queues (%u)\n", RSS should be capitalized in log messages > + vsi->rss_size); > + goto err_free; > + } > + vsi->reconfig_rss = true; > + dev_dbg(&vsi->back->pdev->dev, > + "Reconfigured rss with num_queues (%u)\n", vsi->rss_size); Ditto > + vsi->next_base_queue = num_qps; > + vsi->cnt_q_avail = vsi->num_queue_pairs - num_qps; > + > + /* Update the VSI after updating the VSI queue-mapping > + * information > + */ > + ret = i40e_aq_update_vsi_params(hw, &ctxt, NULL); > + if (ret) { > + dev_info(&pf->pdev->dev, > + "Update vsi tc config failed, err %s aq_err %s\n", > + i40e_stat_str(hw, ret), > + i40e_aq_str(hw, hw->aq.asq_last_status)); > + goto err_free; > + } > + /* update the local VSI info with updated queue map */ > + i40e_vsi_update_queue_map(vsi, &ctxt); > + vsi->info.valid_sections = 0; > + > + /* Create channels for macvlans */ > + INIT_LIST_HEAD(&vsi->macvlan_list); > + vsi->macvlan_cnt = macvlan_cnt; > + for (i = 0; i < macvlan_cnt; i++) { > + ch = kzalloc(sizeof(*ch), GFP_KERNEL); > + if (!ch) { > + ret = -ENOMEM; > + goto err_free; > + } > + INIT_LIST_HEAD(&ch->list); > + ch->num_queue_pairs = qcnt; > + if (!i40e_setup_channel(pf, vsi, ch)) { > + dev_info(&pf->pdev->dev, "Failed to setup macvlan\n"); > + return -EINVAL; Why doesn't this goto err_free? > + } > + ch->parent_vsi = vsi; > + vsi->cnt_q_avail -= ch->num_queue_pairs; > + list_add_tail(&ch->list, &vsi->macvlan_list); > + } > +err_free: Is there anything that should be freed here? > + return ret; > +} > + > +/** > + * i40e_fwd_add - configure macvlans > + * @netdev: net device to configure > + * @vdev: macvlan netdevice > + **/ > +static void *i40e_fwd_add(struct net_device *netdev, struct net_device *vdev) > +{ > + struct i40e_netdev_priv *np = netdev_priv(netdev); > + u16 q_per_macvlan = 0, macvlan_cnt = 0, vectors; > + struct i40e_vsi *vsi = np->vsi; > + struct i40e_pf *pf = vsi->back; > + struct i40e_fwd_adapter *fwd; > + int avail_macvlan, ret; > + > + if ((pf->flags & I40E_FLAG_DCB_ENABLED)) { > + netdev_info(netdev, "Macvlans are not supported when DCB is enabled\n"); > + return ERR_PTR(-EINVAL); > + } > + if ((pf->flags & I40E_FLAG_TC_MQPRIO)) { > + netdev_info(netdev, "Macvlans are not supported when HW TC offload is on\n"); > + return ERR_PTR(-EINVAL); > + } > + > + /* The macvlan device can't be a multiqueue device */ > + if (netif_is_multiqueue(vdev)) > + return ERR_PTR(-ERANGE); > + > + if (!vsi->macvlan_cnt) { > + /* reserve bit 0 for the pf device */ > + set_bit(0, vsi->fwd_bitmask); > + > + /* Try to reserve as many queues for macvlans. First reserve > + * 3/4th of max vectors, then half, then quarter and calculate > + * Qs per macvlan as you go > + */ > + vectors = pf->num_lan_msix; > + if (vectors <= I40E_MAX_MACVLANS && vectors > 96) { > + /* allocate 4 Qs per macvlan and 32 Qs to the PF*/ > + q_per_macvlan = 4; > + macvlan_cnt = (vectors - 32) / 4; > + } else if (vectors <= 96 && vectors > 64) { > + /* allocate 4 Qs per macvlan and 32 Qs to the PF*/ > + q_per_macvlan = 4; > + macvlan_cnt = (vectors - 32) / 4; > + } else if (vectors <= 64 && vectors > 32) { > + /* allocate 2 Qs per macvlan and 16 Qs to the PF*/ > + q_per_macvlan = 2; > + macvlan_cnt = (vectors - 16) / 2; > + } else { > + /* allocate 1 Q per macvlan 16 Qs to the PF*/ > + q_per_macvlan = 1; > + macvlan_cnt = (vectors - 16); > + } > + if (macvlan_cnt == 0) > + return ERR_PTR(-EBUSY); > + > + /* Quiesce VSI queues */ > + i40e_quiesce_vsi(vsi); > + > + /* sets up the macvlans but does not "enable" them */ > + ret = i40e_setup_macvlans(vsi, macvlan_cnt, q_per_macvlan, > + vdev); > + if (ret) > + return ERR_PTR(ret); > + > + /* Unquiesce VSI */ > + i40e_unquiesce_vsi(vsi); > + } > + avail_macvlan = find_first_zero_bit(vsi->fwd_bitmask, > + vsi->macvlan_cnt); > + > + /* create the fwd struct */ > + fwd = kzalloc(sizeof(*fwd), GFP_KERNEL); > + if (!fwd) > + return ERR_PTR(-ENOMEM); > + > + set_bit(avail_macvlan, vsi->fwd_bitmask); > + fwd->bit_no = avail_macvlan; > + netdev_set_sb_channel(vdev, avail_macvlan); > + fwd->netdev = vdev; > + > + if (!netif_running(netdev)) > + return fwd; > + > + /* Set fwd ring up */ > + ret = i40e_fwd_ring_up(vsi, vdev, fwd); > + if (ret) { > + /* unbind the queues and drop the subordinate channel config */ > + netdev_unbind_sb_channel(netdev, vdev); > + netdev_set_sb_channel(vdev, 0); > + > + kfree(fwd); > + return ERR_PTR(-EINVAL); > + } > + return fwd; > +} > + > +/** > + * i40e_reset_ch_rings - Reset the queue contexts in a channel > + * @vsi: the VSI we want to access > + * @ch: the channel we want to access > + */ > +static void i40e_reset_ch_rings(struct i40e_vsi *vsi, struct i40e_channel *ch) > +{ > + struct i40e_ring *tx_ring, *rx_ring; > + u16 pf_q; > + int i; > + > + for (i = 0; i < ch->num_queue_pairs; i++) { > + pf_q = ch->base_queue + i; > + tx_ring = vsi->tx_rings[pf_q]; > + tx_ring->ch = NULL; > + rx_ring = vsi->rx_rings[pf_q]; > + rx_ring->ch = NULL; > + } > +} > + > +/** > + * i40e_del_all_macvlans - Delete all the mac filters on the channels > + * @vsi: the VSI we want to access > + */ > +static void i40e_del_all_macvlans(struct i40e_vsi *vsi) > +{ > + struct i40e_channel *ch, *ch_tmp; > + struct i40e_pf *pf = vsi->back; > + struct i40e_hw *hw = &pf->hw; > + int aq_err, ret = 0; > + > + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, list) { > + if (i40e_is_channel_macvlan(ch)) { > + ret = i40e_del_macvlan_filter(hw, ch->seid, > + i40e_channel_mac(ch), > + &aq_err); > + if (!ret) { > + /* Reset queue contexts */ > + i40e_reset_ch_rings(vsi, ch); > + clear_bit(ch->fwd->bit_no, vsi->fwd_bitmask); > + netdev_unbind_sb_channel(vsi->netdev, > + ch->fwd->netdev); > + netdev_set_sb_channel(ch->fwd->netdev, 0); > + kfree(ch->fwd); > + ch->fwd = NULL; > + } > + } > + } > +} > + > +/** > + * i40e_fwd_del - delete macvlan interfaces > + * @netdev: net device to configure > + * @vdev: macvlan netdevice > + */ > +static void i40e_fwd_del(struct net_device *netdev, void *vdev) > +{ > + struct i40e_netdev_priv *np = netdev_priv(netdev); > + struct i40e_fwd_adapter *fwd = vdev; > + struct i40e_channel *ch, *ch_tmp; > + struct i40e_vsi *vsi = np->vsi; > + struct i40e_pf *pf = vsi->back; > + struct i40e_hw *hw = &pf->hw; > + int aq_err, ret = 0; > + > + /* Find the channel associated with the macvlan and del mac filter */ > + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, list) { > + if (i40e_is_channel_macvlan(ch) && > + ether_addr_equal(i40e_channel_mac(ch), > + fwd->netdev->dev_addr)) { > + ret = i40e_del_macvlan_filter(hw, ch->seid, > + i40e_channel_mac(ch), > + &aq_err); > + if (!ret) { > + /* Reset queue contexts */ > + i40e_reset_ch_rings(vsi, ch); > + clear_bit(ch->fwd->bit_no, vsi->fwd_bitmask); > + netdev_unbind_sb_channel(netdev, fwd->netdev); > + netdev_set_sb_channel(fwd->netdev, 0); > + kfree(ch->fwd); > + ch->fwd = NULL; > + } else { > + dev_info(&pf->pdev->dev, > + "Error deleting mac filter on macvlan err %s, aq_err %s\n", > + i40e_stat_str(hw, ret), > + i40e_aq_str(hw, aq_err)); > + } > + break; > + } > + } > +} > + > /** > * i40e_setup_tc - configure multiple traffic classes > * @netdev: net device to configure > @@ -11580,6 +11993,9 @@ static int i40e_set_features(struct net_device *netdev, > return -EINVAL; > } > > + if (!(features & NETIF_F_HW_L2FW_DOFFLOAD)) > + i40e_del_all_macvlans(vsi); > + > need_reset = i40e_set_ntuple(pf, features); > > if (need_reset) > @@ -12313,6 +12729,8 @@ static const struct net_device_ops i40e_netdev_ops = { > .ndo_bpf = i40e_xdp, > .ndo_xdp_xmit = i40e_xdp_xmit, > .ndo_xsk_async_xmit = i40e_xsk_async_xmit, > + .ndo_dfwd_add_station = i40e_fwd_add, > + .ndo_dfwd_del_station = i40e_fwd_del, > }; > > /** > @@ -12372,6 +12790,9 @@ static int i40e_config_netdev(struct i40e_vsi *vsi) > /* record features VLANs can make use of */ > netdev->vlan_features |= hw_enc_features | NETIF_F_TSO_MANGLEID; > > + /* enable macvlan offloads */ > + netdev->hw_features |= NETIF_F_HW_L2FW_DOFFLOAD; > + > hw_features = hw_enc_features | > NETIF_F_HW_VLAN_CTAG_TX | > NETIF_F_HW_VLAN_CTAG_RX; > -- > 2.17.1 > > _______________________________________________ > Intel-wired-lan mailing list > Intel-wired-lan@osuosl.org > https://lists.osuosl.org/mailman/listinfo/intel-wired-lan
On Wed, 2018-12-26 at 11:01 -0800, Shannon Nelson wrote: > On Tue, Dec 18, 2018 at 1:55 PM Harshitha Ramamurthy > <harshitha.ramamurthy@intel.com> wrote: > > > > This patch enables macvlan offloads on Fortville devices. The idea > > is to use channels as macvlan interfaces. The channels are VSIs of > > type VMDQ. When the first macvlan is created, the maximum no. of > > channels possible are created. From then on, as a macvlan interface > > is created, a mac filter is added to these already created channels > > (VSIs). > > > > This patch builds on top of the recent changes which move > > away from the select_queue implementation of picking the tx queue. > > > > Steps to configure the macvlan: > > 1. sudo ethtool -K ens261f1 l2-fwd-offload on > > 2. ip link add link ens261f1 name macvlan1 type macvlan > > 3. sudo ip link add link ens261f1 name macvlan1 type macvlan > > 4. sudo ip link set macvlan1 up > > Nice to see this finally coming out, thanks. A few nit-pics below, > otherwise > Acked-by: Shannon Nelson <shannon.lee.nelson@gmail.com> Thanks for the review, responses inline to some comments below. Will address the rest of the comments in the next version. > > Cheers, > sln > > > > > Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com > > > > > --- > > drivers/net/ethernet/intel/i40e/i40e.h | 26 ++ > > drivers/net/ethernet/intel/i40e/i40e_main.c | 425 > > +++++++++++++++++++- > > 2 files changed, 449 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/net/ethernet/intel/i40e/i40e.h > > b/drivers/net/ethernet/intel/i40e/i40e.h > > index 4f4de85887a6..7416625e091e 100644 > > --- a/drivers/net/ethernet/intel/i40e/i40e.h > > +++ b/drivers/net/ethernet/intel/i40e/i40e.h > > @@ -27,6 +27,7 @@ > > #include <net/ip6_checksum.h> > > #include <linux/ethtool.h> > > #include <linux/if_vlan.h> > > +#include <linux/if_macvlan.h> > > #include <linux/if_bridge.h> > > #include <linux/clocksource.h> > > #include <linux/net_tstamp.h> > > @@ -390,6 +391,11 @@ struct i40e_flex_pit { > > u8 pit_index; > > }; > > > > +struct i40e_fwd_adapter { > > + struct net_device *netdev; > > + int bit_no; > > +}; > > + > > struct i40e_channel { > > struct list_head list; > > bool initialized; > > @@ -404,11 +410,25 @@ struct i40e_channel { > > struct i40e_aqc_vsi_properties_data info; > > > > u64 max_tx_rate; > > + struct i40e_fwd_adapter *fwd; > > > > /* track this channel belongs to which VSI */ > > struct i40e_vsi *parent_vsi; > > }; > > > > +static inline bool i40e_is_channel_macvlan(struct i40e_channel > > *ch) > > +{ > > + return !!ch->fwd; > > +} > > + > > +static inline u8 *i40e_channel_mac(struct i40e_channel *ch) > > +{ > > + if (i40e_is_channel_macvlan(ch)) > > + return ch->fwd->netdev->dev_addr; > > + else > > + return NULL; > > +} > > + > > /* struct that defines the Ethernet device */ > > struct i40e_pf { > > struct pci_dev *pdev; > > @@ -784,6 +804,12 @@ struct i40e_vsi { > > struct list_head ch_list; > > u16 tc_seid_map[I40E_MAX_TRAFFIC_CLASS]; > > > > + /* macvlan fields */ > > +#define I40E_MAX_MACVLANS 128 /* Max HW capable vectors - 1 > > on FVL */ > > + DECLARE_BITMAP(fwd_bitmask, I40E_MAX_MACVLANS); > > + struct list_head macvlan_list; > > + int macvlan_cnt; > > + > > void *priv; /* client driver data reference. */ > > > > /* VSI specific handlers */ > > diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c > > b/drivers/net/ethernet/intel/i40e/i40e_main.c > > index 1ab1f579343f..1e61c22c27f4 100644 > > --- a/drivers/net/ethernet/intel/i40e/i40e_main.c > > +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c > > @@ -5818,8 +5818,10 @@ static int i40e_add_channel(struct i40e_pf > > *pf, u16 uplink_seid, > > return -ENOENT; > > } > > > > - /* Success, update channel */ > > - ch->enabled_tc = enabled_tc; > > + /* Success, update channel, set enabled_tc only if the > > channel > > + * is not a macvlan > > + */ > > + ch->enabled_tc = !i40e_is_channel_macvlan(ch) && > > enabled_tc; > > ch->seid = ctxt.seid; > > ch->vsi_number = ctxt.vsi_number; > > ch->stat_counter_idx = > > cpu_to_le16(ctxt.info.stat_counter_idx); > > @@ -6811,6 +6813,417 @@ static void > > i40e_vsi_set_default_tc_config(struct i40e_vsi *vsi) > > } > > } > > > > +/** > > + * i40e_del_macvlan_filter > > + * @hw: pointer to the HW structure > > + * @seid: seid of the channel VSI > > + * @macaddr: the mac address to apply as a filter > > + * @aq_err: store the admin Q error > > + * > > + * This function deletes a mac filter on the channel VSI which > > serves as the > > + * macvlan. Returns 0 on success. > > + **/ > > +static i40e_status i40e_del_macvlan_filter(struct i40e_hw *hw, u16 > > seid, > > + const u8 *macaddr, int > > *aq_err) > > +{ > > + struct i40e_aqc_remove_macvlan_element_data element; > > + i40e_status status; > > + > > + memset(&element, 0, sizeof(element)); > > + ether_addr_copy(element.mac_addr, macaddr); > > + element.vlan_tag = 0; > > + element.flags = I40E_AQC_MACVLAN_DEL_PERFECT_MATCH; > > + status = i40e_aq_remove_macvlan(hw, seid, &element, 1, > > NULL); > > + *aq_err = hw->aq.asq_last_status; > > + return status; > > +} > > + > > +/** > > + * i40e_add_macvlan_filter > > + * @hw: pointer to the HW structure > > + * @seid: seid of the channel VSI > > + * @macaddr: the mac address to apply as a filter > > + * @aq_err: store the admin Q error > > + * > > + * This function adds a mac filter on the channel VSI which serves > > as the > > + * macvlan. Returns 0 on success. > > + **/ > > +static i40e_status i40e_add_macvlan_filter(struct i40e_hw *hw, u16 > > seid, > > + const u8 *macaddr, int > > *aq_err) > > +{ > > + struct i40e_aqc_add_macvlan_element_data element; > > + i40e_status status; > > + u16 cmd_flags = 0; > > + > > + ether_addr_copy(element.mac_addr, macaddr); > > + element.vlan_tag = 0; > > + element.queue_number = 0; > > + element.match_method = I40E_AQC_MM_ERR_NO_RES; > > + cmd_flags |= I40E_AQC_MACVLAN_ADD_PERFECT_MATCH; > > + element.flags = cpu_to_le16(cmd_flags); > > + status = i40e_aq_add_macvlan(hw, seid, &element, 1, NULL); > > + *aq_err = hw->aq.asq_last_status; > > + return status; > > +} > > + > > +/** > > + * i40e_fwd_ring_up - bring the macvlan device up > > + * @vsi: the VSI we want to access > > + * @vdev: macvlan netdevice > > + * @fwd: the private fwd structure > > + */ > > +static int i40e_fwd_ring_up(struct i40e_vsi *vsi, struct > > net_device *vdev, > > + struct i40e_fwd_adapter *fwd) > > +{ > > + int ret = 0, num_tc = 1, i, aq_err; > > + struct i40e_channel *ch, *ch_tmp; > > + struct i40e_pf *pf = vsi->back; > > + struct i40e_hw *hw = &pf->hw; > > + > > + if (list_empty(&vsi->macvlan_list)) > > + return -EINVAL; > > + > > + /* Go through the list and find an avaialble channel */ > > s/avaialble/available/ > > > + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, > > list) { > > + if (!i40e_is_channel_macvlan(ch)) { > > Will the channel ever be used for anything else? > Perhaps the sense of this should be to check for in use rather than > is > it a macvlan? So the i40e_is_channel_macvlan macro checks if the channel is in use as a macvlan. The idea is to go through the vsi->macvlan_list and identify the next channel in the list which is available to be handed off to the macvlan device. The macro does this by checking if ch->fwd is valid or not. > > > + ch->fwd = fwd; > > + /* record configuration for macvlan > > interface in vdev */ > > + for (i = 0; i < num_tc; i++) > > + netdev_bind_sb_channel_queue(vsi- > > >netdev, vdev, > > + i, > > + ch- > > >num_queue_pairs, > > + ch- > > >base_queue); > > + for (i = 0; i < ch->num_queue_pairs; i++) { > > + struct i40e_ring *tx_ring, > > *rx_ring; > > + u16 pf_q; > > + > > + pf_q = ch->base_queue + i; > > + > > + /* Get to TX ring ptr */ > > + tx_ring = vsi->tx_rings[pf_q]; > > + tx_ring->ch = ch; > > + > > + /* Get the RX ring ptr */ > > + rx_ring = vsi->rx_rings[pf_q]; > > + rx_ring->ch = ch; > > + } > > + break; > > + } > > + } > > + > > + /* Guarantee all rings are updated before we update the > > + * MAC address filter. > > + */ > > + wmb(); > > + > > + /* Add a mac filter */ > > + ret = i40e_add_macvlan_filter(hw, ch->seid, vdev->dev_addr, > > &aq_err); > > + if (ret) { > > + /* if we cannot add the MAC rule then disable the > > offload */ > > + macvlan_release_l2fw_offload(vdev); > > + for (i = 0; i < ch->num_queue_pairs; i++) { > > + struct i40e_ring *rx_ring; > > + u16 pf_q; > > + > > + pf_q = ch->base_queue + i; > > + rx_ring = vsi->rx_rings[pf_q]; > > + rx_ring->netdev = NULL; > > + } > > + dev_info(&pf->pdev->dev, > > + "Error adding mac filter on macvlan err > > %s, aq_err %s\n", > > + i40e_stat_str(hw, ret), > > + i40e_aq_str(hw, aq_err)); > > + netdev_err(vdev, "L2fwd offload disabled to L2 > > filter error\n"); > > + } > > + return ret; > > +} > > + > > +/** > > + * i40e_setup_macvlans - create the channels which will be > > macvlans > > + * @vsi: the VSI we want to access > > + * @macvlan_cnt: no. of macvlans to be setup > > + * @qcnt: no. of Qs per macvlan > > + * @vdev: macvlan netdevice > > + */ > > +static int i40e_setup_macvlans(struct i40e_vsi *vsi, u16 > > macvlan_cnt, u16 qcnt, > > + struct net_device *vdev) > > +{ > > + struct i40e_pf *pf = vsi->back; > > + struct i40e_hw *hw = &pf->hw; > > + struct i40e_vsi_context ctxt; > > + u16 sections, qmap, num_qps; > > + struct i40e_channel *ch; > > + int i, pow, ret = 0; > > + u8 offset = 0; > > + > > + if (vsi->type != I40E_VSI_MAIN) > > + return -EINVAL; > > + if (!macvlan_cnt) > > + return -EBUSY; > > EBUSY doesn't look right here, maybe EINVAL? > > > + > > + num_qps = vsi->num_queue_pairs - (macvlan_cnt * qcnt); > > + > > + /* find the next higher power-of-2 of num queue pairs */ > > + pow = fls(roundup_pow_of_two(num_qps) - 1); > > + > > + qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) | > > + (pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT); > > + > > + /* Setup context bits for the main VSI */ > > + sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID; > > + sections |= I40E_AQ_VSI_PROP_SCHED_VALID; > > + ctxt.seid = vsi->seid; > > Perhaps memset( 0 ) the ctxt before using it? > > > + ctxt.pf_num = vsi->back->hw.pf_id; > > + ctxt.vf_num = 0; > > + ctxt.uplink_seid = vsi->uplink_seid; > > + ctxt.info = vsi->info; > > + ctxt.info.tc_mapping[0] = cpu_to_le16(qmap); > > + ctxt.info.mapping_flags |= > > cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG); > > + ctxt.info.queue_mapping[0] = cpu_to_le16(vsi->base_queue); > > + ctxt.info.valid_sections |= cpu_to_le16(sections); > > + > > + /* Reconfigure RSS for main VSI with max queue count */ > > s/with max/with new max/ > > > + vsi->rss_size = max_t(u16, num_qps, qcnt); > > + ret = i40e_vsi_config_rss(vsi); > > + if (ret) { > > + dev_info(&vsi->back->pdev->dev, > > + "Failed to reconfig rss for num_queues > > (%u)\n", > > RSS should be capitalized in log messages > > > + vsi->rss_size); > > + goto err_free; > > + } > > + vsi->reconfig_rss = true; > > + dev_dbg(&vsi->back->pdev->dev, > > + "Reconfigured rss with num_queues (%u)\n", vsi- > > >rss_size); > > Ditto > > > + vsi->next_base_queue = num_qps; > > + vsi->cnt_q_avail = vsi->num_queue_pairs - num_qps; > > + > > + /* Update the VSI after updating the VSI queue-mapping > > + * information > > + */ > > + ret = i40e_aq_update_vsi_params(hw, &ctxt, NULL); > > + if (ret) { > > + dev_info(&pf->pdev->dev, > > + "Update vsi tc config failed, err %s > > aq_err %s\n", > > + i40e_stat_str(hw, ret), > > + i40e_aq_str(hw, hw->aq.asq_last_status)); > > + goto err_free; > > + } > > + /* update the local VSI info with updated queue map */ > > + i40e_vsi_update_queue_map(vsi, &ctxt); > > + vsi->info.valid_sections = 0; > > + > > + /* Create channels for macvlans */ > > + INIT_LIST_HEAD(&vsi->macvlan_list); > > + vsi->macvlan_cnt = macvlan_cnt; > > + for (i = 0; i < macvlan_cnt; i++) { > > + ch = kzalloc(sizeof(*ch), GFP_KERNEL); > > + if (!ch) { > > + ret = -ENOMEM; > > + goto err_free; > > + } > > + INIT_LIST_HEAD(&ch->list); > > + ch->num_queue_pairs = qcnt; > > + if (!i40e_setup_channel(pf, vsi, ch)) { > > + dev_info(&pf->pdev->dev, "Failed to setup > > macvlan\n"); > > + return -EINVAL; > > Why doesn't this goto err_free? > > > + } > > + ch->parent_vsi = vsi; > > + vsi->cnt_q_avail -= ch->num_queue_pairs; > > + list_add_tail(&ch->list, &vsi->macvlan_list); > > + } > > +err_free: > > Is there anything that should be freed here? Good point. I guess in this stage where we are setting up all the macvlan VSIs if there is an error, I should free memory and also remove the other channels that have been setup. I will add another function to do that and send the next version along with the other suggested changes. > > > + return ret; > > +} > > + > > +/** > > + * i40e_fwd_add - configure macvlans > > + * @netdev: net device to configure > > + * @vdev: macvlan netdevice > > + **/ > > +static void *i40e_fwd_add(struct net_device *netdev, struct > > net_device *vdev) > > +{ > > + struct i40e_netdev_priv *np = netdev_priv(netdev); > > + u16 q_per_macvlan = 0, macvlan_cnt = 0, vectors; > > + struct i40e_vsi *vsi = np->vsi; > > + struct i40e_pf *pf = vsi->back; > > + struct i40e_fwd_adapter *fwd; > > + int avail_macvlan, ret; > > + > > + if ((pf->flags & I40E_FLAG_DCB_ENABLED)) { > > + netdev_info(netdev, "Macvlans are not supported > > when DCB is enabled\n"); > > + return ERR_PTR(-EINVAL); > > + } > > + if ((pf->flags & I40E_FLAG_TC_MQPRIO)) { > > + netdev_info(netdev, "Macvlans are not supported > > when HW TC offload is on\n"); > > + return ERR_PTR(-EINVAL); > > + } > > + > > + /* The macvlan device can't be a multiqueue device */ > > + if (netif_is_multiqueue(vdev)) > > + return ERR_PTR(-ERANGE); > > + > > + if (!vsi->macvlan_cnt) { > > + /* reserve bit 0 for the pf device */ > > + set_bit(0, vsi->fwd_bitmask); > > + > > + /* Try to reserve as many queues for macvlans. > > First reserve > > + * 3/4th of max vectors, then half, then quarter > > and calculate > > + * Qs per macvlan as you go > > + */ > > + vectors = pf->num_lan_msix; > > + if (vectors <= I40E_MAX_MACVLANS && vectors > 96) { > > + /* allocate 4 Qs per macvlan and 32 Qs to > > the PF*/ > > + q_per_macvlan = 4; > > + macvlan_cnt = (vectors - 32) / 4; > > + } else if (vectors <= 96 && vectors > 64) { > > + /* allocate 4 Qs per macvlan and 32 Qs to > > the PF*/ > > + q_per_macvlan = 4; > > + macvlan_cnt = (vectors - 32) / 4; > > + } else if (vectors <= 64 && vectors > 32) { > > + /* allocate 2 Qs per macvlan and 16 Qs to > > the PF*/ > > + q_per_macvlan = 2; > > + macvlan_cnt = (vectors - 16) / 2; > > + } else { > > + /* allocate 1 Q per macvlan 16 Qs to the > > PF*/ > > + q_per_macvlan = 1; > > + macvlan_cnt = (vectors - 16); > > + } > > + if (macvlan_cnt == 0) > > + return ERR_PTR(-EBUSY); > > + > > + /* Quiesce VSI queues */ > > + i40e_quiesce_vsi(vsi); > > + > > + /* sets up the macvlans but does not "enable" them > > */ > > + ret = i40e_setup_macvlans(vsi, macvlan_cnt, > > q_per_macvlan, > > + vdev); > > + if (ret) > > + return ERR_PTR(ret); > > + > > + /* Unquiesce VSI */ > > + i40e_unquiesce_vsi(vsi); > > + } > > + avail_macvlan = find_first_zero_bit(vsi->fwd_bitmask, > > + vsi->macvlan_cnt); > > + > > + /* create the fwd struct */ > > + fwd = kzalloc(sizeof(*fwd), GFP_KERNEL); > > + if (!fwd) > > + return ERR_PTR(-ENOMEM); > > + > > + set_bit(avail_macvlan, vsi->fwd_bitmask); > > + fwd->bit_no = avail_macvlan; > > + netdev_set_sb_channel(vdev, avail_macvlan); > > + fwd->netdev = vdev; > > + > > + if (!netif_running(netdev)) > > + return fwd; > > + > > + /* Set fwd ring up */ > > + ret = i40e_fwd_ring_up(vsi, vdev, fwd); > > + if (ret) { > > + /* unbind the queues and drop the subordinate > > channel config */ > > + netdev_unbind_sb_channel(netdev, vdev); > > + netdev_set_sb_channel(vdev, 0); > > + > > + kfree(fwd); > > + return ERR_PTR(-EINVAL); > > + } > > + return fwd; > > +} > > + > > +/** > > + * i40e_reset_ch_rings - Reset the queue contexts in a channel > > + * @vsi: the VSI we want to access > > + * @ch: the channel we want to access > > + */ > > +static void i40e_reset_ch_rings(struct i40e_vsi *vsi, struct > > i40e_channel *ch) > > +{ > > + struct i40e_ring *tx_ring, *rx_ring; > > + u16 pf_q; > > + int i; > > + > > + for (i = 0; i < ch->num_queue_pairs; i++) { > > + pf_q = ch->base_queue + i; > > + tx_ring = vsi->tx_rings[pf_q]; > > + tx_ring->ch = NULL; > > + rx_ring = vsi->rx_rings[pf_q]; > > + rx_ring->ch = NULL; > > + } > > +} > > + > > +/** > > + * i40e_del_all_macvlans - Delete all the mac filters on the > > channels > > + * @vsi: the VSI we want to access > > + */ > > +static void i40e_del_all_macvlans(struct i40e_vsi *vsi) > > +{ > > + struct i40e_channel *ch, *ch_tmp; > > + struct i40e_pf *pf = vsi->back; > > + struct i40e_hw *hw = &pf->hw; > > + int aq_err, ret = 0; > > + > > + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, > > list) { > > + if (i40e_is_channel_macvlan(ch)) { > > + ret = i40e_del_macvlan_filter(hw, ch->seid, > > + i40e_channel_ > > mac(ch), > > + &aq_err); > > + if (!ret) { > > + /* Reset queue contexts */ > > + i40e_reset_ch_rings(vsi, ch); > > + clear_bit(ch->fwd->bit_no, vsi- > > >fwd_bitmask); > > + netdev_unbind_sb_channel(vsi- > > >netdev, > > + ch->fwd- > > >netdev); > > + netdev_set_sb_channel(ch->fwd- > > >netdev, 0); > > + kfree(ch->fwd); > > + ch->fwd = NULL; > > + } > > + } > > + } > > +} > > + > > +/** > > + * i40e_fwd_del - delete macvlan interfaces > > + * @netdev: net device to configure > > + * @vdev: macvlan netdevice > > + */ > > +static void i40e_fwd_del(struct net_device *netdev, void *vdev) > > +{ > > + struct i40e_netdev_priv *np = netdev_priv(netdev); > > + struct i40e_fwd_adapter *fwd = vdev; > > + struct i40e_channel *ch, *ch_tmp; > > + struct i40e_vsi *vsi = np->vsi; > > + struct i40e_pf *pf = vsi->back; > > + struct i40e_hw *hw = &pf->hw; > > + int aq_err, ret = 0; > > + > > + /* Find the channel associated with the macvlan and del mac > > filter */ > > + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, > > list) { > > + if (i40e_is_channel_macvlan(ch) && > > + ether_addr_equal(i40e_channel_mac(ch), > > + fwd->netdev->dev_addr)) { > > + ret = i40e_del_macvlan_filter(hw, ch->seid, > > + i40e_channel_ > > mac(ch), > > + &aq_err); > > + if (!ret) { > > + /* Reset queue contexts */ > > + i40e_reset_ch_rings(vsi, ch); > > + clear_bit(ch->fwd->bit_no, vsi- > > >fwd_bitmask); > > + netdev_unbind_sb_channel(netdev, > > fwd->netdev); > > + netdev_set_sb_channel(fwd->netdev, > > 0); > > + kfree(ch->fwd); > > + ch->fwd = NULL; > > + } else { > > + dev_info(&pf->pdev->dev, > > + "Error deleting > > mac filter on macvlan err %s, aq_err %s\n", > > + i40e_stat_str(hw, > > ret), > > + i40e_aq_str(hw, > > aq_err)); > > + } > > + break; > > + } > > + } > > +} > > + > > /** > > * i40e_setup_tc - configure multiple traffic classes > > * @netdev: net device to configure > > @@ -11580,6 +11993,9 @@ static int i40e_set_features(struct > > net_device *netdev, > > return -EINVAL; > > } > > > > + if (!(features & NETIF_F_HW_L2FW_DOFFLOAD)) > > + i40e_del_all_macvlans(vsi); > > + > > need_reset = i40e_set_ntuple(pf, features); > > > > if (need_reset) > > @@ -12313,6 +12729,8 @@ static const struct net_device_ops > > i40e_netdev_ops = { > > .ndo_bpf = i40e_xdp, > > .ndo_xdp_xmit = i40e_xdp_xmit, > > .ndo_xsk_async_xmit = i40e_xsk_async_xmit, > > + .ndo_dfwd_add_station = i40e_fwd_add, > > + .ndo_dfwd_del_station = i40e_fwd_del, > > }; > > > > /** > > @@ -12372,6 +12790,9 @@ static int i40e_config_netdev(struct > > i40e_vsi *vsi) > > /* record features VLANs can make use of */ > > netdev->vlan_features |= hw_enc_features | > > NETIF_F_TSO_MANGLEID; > > > > + /* enable macvlan offloads */ > > + netdev->hw_features |= NETIF_F_HW_L2FW_DOFFLOAD; > > + > > hw_features = hw_enc_features | > > NETIF_F_HW_VLAN_CTAG_TX | > > NETIF_F_HW_VLAN_CTAG_RX; > > -- > > 2.17.1 > > > > _______________________________________________ > > Intel-wired-lan mailing list > > Intel-wired-lan@osuosl.org > > https://lists.osuosl.org/mailman/listinfo/intel-wired-lan > > >
diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h index 4f4de85887a6..7416625e091e 100644 --- a/drivers/net/ethernet/intel/i40e/i40e.h +++ b/drivers/net/ethernet/intel/i40e/i40e.h @@ -27,6 +27,7 @@ #include <net/ip6_checksum.h> #include <linux/ethtool.h> #include <linux/if_vlan.h> +#include <linux/if_macvlan.h> #include <linux/if_bridge.h> #include <linux/clocksource.h> #include <linux/net_tstamp.h> @@ -390,6 +391,11 @@ struct i40e_flex_pit { u8 pit_index; }; +struct i40e_fwd_adapter { + struct net_device *netdev; + int bit_no; +}; + struct i40e_channel { struct list_head list; bool initialized; @@ -404,11 +410,25 @@ struct i40e_channel { struct i40e_aqc_vsi_properties_data info; u64 max_tx_rate; + struct i40e_fwd_adapter *fwd; /* track this channel belongs to which VSI */ struct i40e_vsi *parent_vsi; }; +static inline bool i40e_is_channel_macvlan(struct i40e_channel *ch) +{ + return !!ch->fwd; +} + +static inline u8 *i40e_channel_mac(struct i40e_channel *ch) +{ + if (i40e_is_channel_macvlan(ch)) + return ch->fwd->netdev->dev_addr; + else + return NULL; +} + /* struct that defines the Ethernet device */ struct i40e_pf { struct pci_dev *pdev; @@ -784,6 +804,12 @@ struct i40e_vsi { struct list_head ch_list; u16 tc_seid_map[I40E_MAX_TRAFFIC_CLASS]; + /* macvlan fields */ +#define I40E_MAX_MACVLANS 128 /* Max HW capable vectors - 1 on FVL */ + DECLARE_BITMAP(fwd_bitmask, I40E_MAX_MACVLANS); + struct list_head macvlan_list; + int macvlan_cnt; + void *priv; /* client driver data reference. */ /* VSI specific handlers */ diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index 1ab1f579343f..1e61c22c27f4 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -5818,8 +5818,10 @@ static int i40e_add_channel(struct i40e_pf *pf, u16 uplink_seid, return -ENOENT; } - /* Success, update channel */ - ch->enabled_tc = enabled_tc; + /* Success, update channel, set enabled_tc only if the channel + * is not a macvlan + */ + ch->enabled_tc = !i40e_is_channel_macvlan(ch) && enabled_tc; ch->seid = ctxt.seid; ch->vsi_number = ctxt.vsi_number; ch->stat_counter_idx = cpu_to_le16(ctxt.info.stat_counter_idx); @@ -6811,6 +6813,417 @@ static void i40e_vsi_set_default_tc_config(struct i40e_vsi *vsi) } } +/** + * i40e_del_macvlan_filter + * @hw: pointer to the HW structure + * @seid: seid of the channel VSI + * @macaddr: the mac address to apply as a filter + * @aq_err: store the admin Q error + * + * This function deletes a mac filter on the channel VSI which serves as the + * macvlan. Returns 0 on success. + **/ +static i40e_status i40e_del_macvlan_filter(struct i40e_hw *hw, u16 seid, + const u8 *macaddr, int *aq_err) +{ + struct i40e_aqc_remove_macvlan_element_data element; + i40e_status status; + + memset(&element, 0, sizeof(element)); + ether_addr_copy(element.mac_addr, macaddr); + element.vlan_tag = 0; + element.flags = I40E_AQC_MACVLAN_DEL_PERFECT_MATCH; + status = i40e_aq_remove_macvlan(hw, seid, &element, 1, NULL); + *aq_err = hw->aq.asq_last_status; + return status; +} + +/** + * i40e_add_macvlan_filter + * @hw: pointer to the HW structure + * @seid: seid of the channel VSI + * @macaddr: the mac address to apply as a filter + * @aq_err: store the admin Q error + * + * This function adds a mac filter on the channel VSI which serves as the + * macvlan. Returns 0 on success. + **/ +static i40e_status i40e_add_macvlan_filter(struct i40e_hw *hw, u16 seid, + const u8 *macaddr, int *aq_err) +{ + struct i40e_aqc_add_macvlan_element_data element; + i40e_status status; + u16 cmd_flags = 0; + + ether_addr_copy(element.mac_addr, macaddr); + element.vlan_tag = 0; + element.queue_number = 0; + element.match_method = I40E_AQC_MM_ERR_NO_RES; + cmd_flags |= I40E_AQC_MACVLAN_ADD_PERFECT_MATCH; + element.flags = cpu_to_le16(cmd_flags); + status = i40e_aq_add_macvlan(hw, seid, &element, 1, NULL); + *aq_err = hw->aq.asq_last_status; + return status; +} + +/** + * i40e_fwd_ring_up - bring the macvlan device up + * @vsi: the VSI we want to access + * @vdev: macvlan netdevice + * @fwd: the private fwd structure + */ +static int i40e_fwd_ring_up(struct i40e_vsi *vsi, struct net_device *vdev, + struct i40e_fwd_adapter *fwd) +{ + int ret = 0, num_tc = 1, i, aq_err; + struct i40e_channel *ch, *ch_tmp; + struct i40e_pf *pf = vsi->back; + struct i40e_hw *hw = &pf->hw; + + if (list_empty(&vsi->macvlan_list)) + return -EINVAL; + + /* Go through the list and find an avaialble channel */ + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, list) { + if (!i40e_is_channel_macvlan(ch)) { + ch->fwd = fwd; + /* record configuration for macvlan interface in vdev */ + for (i = 0; i < num_tc; i++) + netdev_bind_sb_channel_queue(vsi->netdev, vdev, + i, + ch->num_queue_pairs, + ch->base_queue); + for (i = 0; i < ch->num_queue_pairs; i++) { + struct i40e_ring *tx_ring, *rx_ring; + u16 pf_q; + + pf_q = ch->base_queue + i; + + /* Get to TX ring ptr */ + tx_ring = vsi->tx_rings[pf_q]; + tx_ring->ch = ch; + + /* Get the RX ring ptr */ + rx_ring = vsi->rx_rings[pf_q]; + rx_ring->ch = ch; + } + break; + } + } + + /* Guarantee all rings are updated before we update the + * MAC address filter. + */ + wmb(); + + /* Add a mac filter */ + ret = i40e_add_macvlan_filter(hw, ch->seid, vdev->dev_addr, &aq_err); + if (ret) { + /* if we cannot add the MAC rule then disable the offload */ + macvlan_release_l2fw_offload(vdev); + for (i = 0; i < ch->num_queue_pairs; i++) { + struct i40e_ring *rx_ring; + u16 pf_q; + + pf_q = ch->base_queue + i; + rx_ring = vsi->rx_rings[pf_q]; + rx_ring->netdev = NULL; + } + dev_info(&pf->pdev->dev, + "Error adding mac filter on macvlan err %s, aq_err %s\n", + i40e_stat_str(hw, ret), + i40e_aq_str(hw, aq_err)); + netdev_err(vdev, "L2fwd offload disabled to L2 filter error\n"); + } + return ret; +} + +/** + * i40e_setup_macvlans - create the channels which will be macvlans + * @vsi: the VSI we want to access + * @macvlan_cnt: no. of macvlans to be setup + * @qcnt: no. of Qs per macvlan + * @vdev: macvlan netdevice + */ +static int i40e_setup_macvlans(struct i40e_vsi *vsi, u16 macvlan_cnt, u16 qcnt, + struct net_device *vdev) +{ + struct i40e_pf *pf = vsi->back; + struct i40e_hw *hw = &pf->hw; + struct i40e_vsi_context ctxt; + u16 sections, qmap, num_qps; + struct i40e_channel *ch; + int i, pow, ret = 0; + u8 offset = 0; + + if (vsi->type != I40E_VSI_MAIN) + return -EINVAL; + if (!macvlan_cnt) + return -EBUSY; + + num_qps = vsi->num_queue_pairs - (macvlan_cnt * qcnt); + + /* find the next higher power-of-2 of num queue pairs */ + pow = fls(roundup_pow_of_two(num_qps) - 1); + + qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) | + (pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT); + + /* Setup context bits for the main VSI */ + sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID; + sections |= I40E_AQ_VSI_PROP_SCHED_VALID; + ctxt.seid = vsi->seid; + ctxt.pf_num = vsi->back->hw.pf_id; + ctxt.vf_num = 0; + ctxt.uplink_seid = vsi->uplink_seid; + ctxt.info = vsi->info; + ctxt.info.tc_mapping[0] = cpu_to_le16(qmap); + ctxt.info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG); + ctxt.info.queue_mapping[0] = cpu_to_le16(vsi->base_queue); + ctxt.info.valid_sections |= cpu_to_le16(sections); + + /* Reconfigure RSS for main VSI with max queue count */ + vsi->rss_size = max_t(u16, num_qps, qcnt); + ret = i40e_vsi_config_rss(vsi); + if (ret) { + dev_info(&vsi->back->pdev->dev, + "Failed to reconfig rss for num_queues (%u)\n", + vsi->rss_size); + goto err_free; + } + vsi->reconfig_rss = true; + dev_dbg(&vsi->back->pdev->dev, + "Reconfigured rss with num_queues (%u)\n", vsi->rss_size); + vsi->next_base_queue = num_qps; + vsi->cnt_q_avail = vsi->num_queue_pairs - num_qps; + + /* Update the VSI after updating the VSI queue-mapping + * information + */ + ret = i40e_aq_update_vsi_params(hw, &ctxt, NULL); + if (ret) { + dev_info(&pf->pdev->dev, + "Update vsi tc config failed, err %s aq_err %s\n", + i40e_stat_str(hw, ret), + i40e_aq_str(hw, hw->aq.asq_last_status)); + goto err_free; + } + /* update the local VSI info with updated queue map */ + i40e_vsi_update_queue_map(vsi, &ctxt); + vsi->info.valid_sections = 0; + + /* Create channels for macvlans */ + INIT_LIST_HEAD(&vsi->macvlan_list); + vsi->macvlan_cnt = macvlan_cnt; + for (i = 0; i < macvlan_cnt; i++) { + ch = kzalloc(sizeof(*ch), GFP_KERNEL); + if (!ch) { + ret = -ENOMEM; + goto err_free; + } + INIT_LIST_HEAD(&ch->list); + ch->num_queue_pairs = qcnt; + if (!i40e_setup_channel(pf, vsi, ch)) { + dev_info(&pf->pdev->dev, "Failed to setup macvlan\n"); + return -EINVAL; + } + ch->parent_vsi = vsi; + vsi->cnt_q_avail -= ch->num_queue_pairs; + list_add_tail(&ch->list, &vsi->macvlan_list); + } +err_free: + return ret; +} + +/** + * i40e_fwd_add - configure macvlans + * @netdev: net device to configure + * @vdev: macvlan netdevice + **/ +static void *i40e_fwd_add(struct net_device *netdev, struct net_device *vdev) +{ + struct i40e_netdev_priv *np = netdev_priv(netdev); + u16 q_per_macvlan = 0, macvlan_cnt = 0, vectors; + struct i40e_vsi *vsi = np->vsi; + struct i40e_pf *pf = vsi->back; + struct i40e_fwd_adapter *fwd; + int avail_macvlan, ret; + + if ((pf->flags & I40E_FLAG_DCB_ENABLED)) { + netdev_info(netdev, "Macvlans are not supported when DCB is enabled\n"); + return ERR_PTR(-EINVAL); + } + if ((pf->flags & I40E_FLAG_TC_MQPRIO)) { + netdev_info(netdev, "Macvlans are not supported when HW TC offload is on\n"); + return ERR_PTR(-EINVAL); + } + + /* The macvlan device can't be a multiqueue device */ + if (netif_is_multiqueue(vdev)) + return ERR_PTR(-ERANGE); + + if (!vsi->macvlan_cnt) { + /* reserve bit 0 for the pf device */ + set_bit(0, vsi->fwd_bitmask); + + /* Try to reserve as many queues for macvlans. First reserve + * 3/4th of max vectors, then half, then quarter and calculate + * Qs per macvlan as you go + */ + vectors = pf->num_lan_msix; + if (vectors <= I40E_MAX_MACVLANS && vectors > 96) { + /* allocate 4 Qs per macvlan and 32 Qs to the PF*/ + q_per_macvlan = 4; + macvlan_cnt = (vectors - 32) / 4; + } else if (vectors <= 96 && vectors > 64) { + /* allocate 4 Qs per macvlan and 32 Qs to the PF*/ + q_per_macvlan = 4; + macvlan_cnt = (vectors - 32) / 4; + } else if (vectors <= 64 && vectors > 32) { + /* allocate 2 Qs per macvlan and 16 Qs to the PF*/ + q_per_macvlan = 2; + macvlan_cnt = (vectors - 16) / 2; + } else { + /* allocate 1 Q per macvlan 16 Qs to the PF*/ + q_per_macvlan = 1; + macvlan_cnt = (vectors - 16); + } + if (macvlan_cnt == 0) + return ERR_PTR(-EBUSY); + + /* Quiesce VSI queues */ + i40e_quiesce_vsi(vsi); + + /* sets up the macvlans but does not "enable" them */ + ret = i40e_setup_macvlans(vsi, macvlan_cnt, q_per_macvlan, + vdev); + if (ret) + return ERR_PTR(ret); + + /* Unquiesce VSI */ + i40e_unquiesce_vsi(vsi); + } + avail_macvlan = find_first_zero_bit(vsi->fwd_bitmask, + vsi->macvlan_cnt); + + /* create the fwd struct */ + fwd = kzalloc(sizeof(*fwd), GFP_KERNEL); + if (!fwd) + return ERR_PTR(-ENOMEM); + + set_bit(avail_macvlan, vsi->fwd_bitmask); + fwd->bit_no = avail_macvlan; + netdev_set_sb_channel(vdev, avail_macvlan); + fwd->netdev = vdev; + + if (!netif_running(netdev)) + return fwd; + + /* Set fwd ring up */ + ret = i40e_fwd_ring_up(vsi, vdev, fwd); + if (ret) { + /* unbind the queues and drop the subordinate channel config */ + netdev_unbind_sb_channel(netdev, vdev); + netdev_set_sb_channel(vdev, 0); + + kfree(fwd); + return ERR_PTR(-EINVAL); + } + return fwd; +} + +/** + * i40e_reset_ch_rings - Reset the queue contexts in a channel + * @vsi: the VSI we want to access + * @ch: the channel we want to access + */ +static void i40e_reset_ch_rings(struct i40e_vsi *vsi, struct i40e_channel *ch) +{ + struct i40e_ring *tx_ring, *rx_ring; + u16 pf_q; + int i; + + for (i = 0; i < ch->num_queue_pairs; i++) { + pf_q = ch->base_queue + i; + tx_ring = vsi->tx_rings[pf_q]; + tx_ring->ch = NULL; + rx_ring = vsi->rx_rings[pf_q]; + rx_ring->ch = NULL; + } +} + +/** + * i40e_del_all_macvlans - Delete all the mac filters on the channels + * @vsi: the VSI we want to access + */ +static void i40e_del_all_macvlans(struct i40e_vsi *vsi) +{ + struct i40e_channel *ch, *ch_tmp; + struct i40e_pf *pf = vsi->back; + struct i40e_hw *hw = &pf->hw; + int aq_err, ret = 0; + + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, list) { + if (i40e_is_channel_macvlan(ch)) { + ret = i40e_del_macvlan_filter(hw, ch->seid, + i40e_channel_mac(ch), + &aq_err); + if (!ret) { + /* Reset queue contexts */ + i40e_reset_ch_rings(vsi, ch); + clear_bit(ch->fwd->bit_no, vsi->fwd_bitmask); + netdev_unbind_sb_channel(vsi->netdev, + ch->fwd->netdev); + netdev_set_sb_channel(ch->fwd->netdev, 0); + kfree(ch->fwd); + ch->fwd = NULL; + } + } + } +} + +/** + * i40e_fwd_del - delete macvlan interfaces + * @netdev: net device to configure + * @vdev: macvlan netdevice + */ +static void i40e_fwd_del(struct net_device *netdev, void *vdev) +{ + struct i40e_netdev_priv *np = netdev_priv(netdev); + struct i40e_fwd_adapter *fwd = vdev; + struct i40e_channel *ch, *ch_tmp; + struct i40e_vsi *vsi = np->vsi; + struct i40e_pf *pf = vsi->back; + struct i40e_hw *hw = &pf->hw; + int aq_err, ret = 0; + + /* Find the channel associated with the macvlan and del mac filter */ + list_for_each_entry_safe(ch, ch_tmp, &vsi->macvlan_list, list) { + if (i40e_is_channel_macvlan(ch) && + ether_addr_equal(i40e_channel_mac(ch), + fwd->netdev->dev_addr)) { + ret = i40e_del_macvlan_filter(hw, ch->seid, + i40e_channel_mac(ch), + &aq_err); + if (!ret) { + /* Reset queue contexts */ + i40e_reset_ch_rings(vsi, ch); + clear_bit(ch->fwd->bit_no, vsi->fwd_bitmask); + netdev_unbind_sb_channel(netdev, fwd->netdev); + netdev_set_sb_channel(fwd->netdev, 0); + kfree(ch->fwd); + ch->fwd = NULL; + } else { + dev_info(&pf->pdev->dev, + "Error deleting mac filter on macvlan err %s, aq_err %s\n", + i40e_stat_str(hw, ret), + i40e_aq_str(hw, aq_err)); + } + break; + } + } +} + /** * i40e_setup_tc - configure multiple traffic classes * @netdev: net device to configure @@ -11580,6 +11993,9 @@ static int i40e_set_features(struct net_device *netdev, return -EINVAL; } + if (!(features & NETIF_F_HW_L2FW_DOFFLOAD)) + i40e_del_all_macvlans(vsi); + need_reset = i40e_set_ntuple(pf, features); if (need_reset) @@ -12313,6 +12729,8 @@ static const struct net_device_ops i40e_netdev_ops = { .ndo_bpf = i40e_xdp, .ndo_xdp_xmit = i40e_xdp_xmit, .ndo_xsk_async_xmit = i40e_xsk_async_xmit, + .ndo_dfwd_add_station = i40e_fwd_add, + .ndo_dfwd_del_station = i40e_fwd_del, }; /** @@ -12372,6 +12790,9 @@ static int i40e_config_netdev(struct i40e_vsi *vsi) /* record features VLANs can make use of */ netdev->vlan_features |= hw_enc_features | NETIF_F_TSO_MANGLEID; + /* enable macvlan offloads */ + netdev->hw_features |= NETIF_F_HW_L2FW_DOFFLOAD; + hw_features = hw_enc_features | NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
This patch enables macvlan offloads on Fortville devices. The idea is to use channels as macvlan interfaces. The channels are VSIs of type VMDQ. When the first macvlan is created, the maximum no. of channels possible are created. From then on, as a macvlan interface is created, a mac filter is added to these already created channels (VSIs). This patch builds on top of the recent changes which move away from the select_queue implementation of picking the tx queue. Steps to configure the macvlan: 1. sudo ethtool -K ens261f1 l2-fwd-offload on 2. ip link add link ens261f1 name macvlan1 type macvlan 3. sudo ip link add link ens261f1 name macvlan1 type macvlan 4. sudo ip link set macvlan1 up Signed-off-by: Harshitha Ramamurthy <harshitha.ramamurthy@intel.com> --- drivers/net/ethernet/intel/i40e/i40e.h | 26 ++ drivers/net/ethernet/intel/i40e/i40e_main.c | 425 +++++++++++++++++++- 2 files changed, 449 insertions(+), 2 deletions(-)