Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / intel / i40e / i40e_virtchnl_pf.c
diff --git a/kernel/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c b/kernel/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c
new file mode 100644 (file)
index 0000000..4e9376d
--- /dev/null
@@ -0,0 +1,2402 @@
+/*******************************************************************************
+ *
+ * Intel Ethernet Controller XL710 Family Linux Driver
+ * Copyright(c) 2013 - 2015 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ * Contact Information:
+ * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
+ * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
+ *
+ ******************************************************************************/
+
+#include "i40e.h"
+
+/*********************notification routines***********************/
+
+/**
+ * i40e_vc_vf_broadcast
+ * @pf: pointer to the PF structure
+ * @opcode: operation code
+ * @retval: return value
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * send a message to all VFs on a given PF
+ **/
+static void i40e_vc_vf_broadcast(struct i40e_pf *pf,
+                                enum i40e_virtchnl_ops v_opcode,
+                                i40e_status v_retval, u8 *msg,
+                                u16 msglen)
+{
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_vf *vf = pf->vf;
+       int i;
+
+       for (i = 0; i < pf->num_alloc_vfs; i++, vf++) {
+               int abs_vf_id = vf->vf_id + hw->func_caps.vf_base_id;
+               /* Not all vfs are enabled so skip the ones that are not */
+               if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states) &&
+                   !test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states))
+                       continue;
+
+               /* Ignore return value on purpose - a given VF may fail, but
+                * we need to keep going and send to all of them
+                */
+               i40e_aq_send_msg_to_vf(hw, abs_vf_id, v_opcode, v_retval,
+                                      msg, msglen, NULL);
+       }
+}
+
+/**
+ * i40e_vc_notify_link_state
+ * @vf: pointer to the VF structure
+ *
+ * send a link status message to a single VF
+ **/
+static void i40e_vc_notify_vf_link_state(struct i40e_vf *vf)
+{
+       struct i40e_virtchnl_pf_event pfe;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_link_status *ls = &pf->hw.phy.link_info;
+       int abs_vf_id = vf->vf_id + hw->func_caps.vf_base_id;
+
+       pfe.event = I40E_VIRTCHNL_EVENT_LINK_CHANGE;
+       pfe.severity = I40E_PF_EVENT_SEVERITY_INFO;
+       if (vf->link_forced) {
+               pfe.event_data.link_event.link_status = vf->link_up;
+               pfe.event_data.link_event.link_speed =
+                       (vf->link_up ? I40E_LINK_SPEED_40GB : 0);
+       } else {
+               pfe.event_data.link_event.link_status =
+                       ls->link_info & I40E_AQ_LINK_UP;
+               pfe.event_data.link_event.link_speed = ls->link_speed;
+       }
+       i40e_aq_send_msg_to_vf(hw, abs_vf_id, I40E_VIRTCHNL_OP_EVENT,
+                              0, (u8 *)&pfe, sizeof(pfe), NULL);
+}
+
+/**
+ * i40e_vc_notify_link_state
+ * @pf: pointer to the PF structure
+ *
+ * send a link status message to all VFs on a given PF
+ **/
+void i40e_vc_notify_link_state(struct i40e_pf *pf)
+{
+       int i;
+
+       for (i = 0; i < pf->num_alloc_vfs; i++)
+               i40e_vc_notify_vf_link_state(&pf->vf[i]);
+}
+
+/**
+ * i40e_vc_notify_reset
+ * @pf: pointer to the PF structure
+ *
+ * indicate a pending reset to all VFs on a given PF
+ **/
+void i40e_vc_notify_reset(struct i40e_pf *pf)
+{
+       struct i40e_virtchnl_pf_event pfe;
+
+       pfe.event = I40E_VIRTCHNL_EVENT_RESET_IMPENDING;
+       pfe.severity = I40E_PF_EVENT_SEVERITY_CERTAIN_DOOM;
+       i40e_vc_vf_broadcast(pf, I40E_VIRTCHNL_OP_EVENT, 0,
+                            (u8 *)&pfe, sizeof(struct i40e_virtchnl_pf_event));
+}
+
+/**
+ * i40e_vc_notify_vf_reset
+ * @vf: pointer to the VF structure
+ *
+ * indicate a pending reset to the given VF
+ **/
+void i40e_vc_notify_vf_reset(struct i40e_vf *vf)
+{
+       struct i40e_virtchnl_pf_event pfe;
+       int abs_vf_id;
+
+       /* validate the request */
+       if (!vf || vf->vf_id >= vf->pf->num_alloc_vfs)
+               return;
+
+       /* verify if the VF is in either init or active before proceeding */
+       if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states) &&
+           !test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states))
+               return;
+
+       abs_vf_id = vf->vf_id + vf->pf->hw.func_caps.vf_base_id;
+
+       pfe.event = I40E_VIRTCHNL_EVENT_RESET_IMPENDING;
+       pfe.severity = I40E_PF_EVENT_SEVERITY_CERTAIN_DOOM;
+       i40e_aq_send_msg_to_vf(&vf->pf->hw, abs_vf_id, I40E_VIRTCHNL_OP_EVENT,
+                              0, (u8 *)&pfe,
+                              sizeof(struct i40e_virtchnl_pf_event), NULL);
+}
+/***********************misc routines*****************************/
+
+/**
+ * i40e_vc_disable_vf
+ * @pf: pointer to the PF info
+ * @vf: pointer to the VF info
+ *
+ * Disable the VF through a SW reset
+ **/
+static inline void i40e_vc_disable_vf(struct i40e_pf *pf, struct i40e_vf *vf)
+{
+       struct i40e_hw *hw = &pf->hw;
+       u32 reg;
+
+       reg = rd32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id));
+       reg |= I40E_VPGEN_VFRTRIG_VFSWR_MASK;
+       wr32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id), reg);
+       i40e_flush(hw);
+}
+
+/**
+ * i40e_vc_isvalid_vsi_id
+ * @vf: pointer to the VF info
+ * @vsi_id: VF relative VSI id
+ *
+ * check for the valid VSI id
+ **/
+static inline bool i40e_vc_isvalid_vsi_id(struct i40e_vf *vf, u16 vsi_id)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = i40e_find_vsi_from_id(pf, vsi_id);
+
+       return (vsi && (vsi->vf_id == vf->vf_id));
+}
+
+/**
+ * i40e_vc_isvalid_queue_id
+ * @vf: pointer to the VF info
+ * @vsi_id: vsi id
+ * @qid: vsi relative queue id
+ *
+ * check for the valid queue id
+ **/
+static inline bool i40e_vc_isvalid_queue_id(struct i40e_vf *vf, u16 vsi_id,
+                                           u8 qid)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = i40e_find_vsi_from_id(pf, vsi_id);
+
+       return (vsi && (qid < vsi->alloc_queue_pairs));
+}
+
+/**
+ * i40e_vc_isvalid_vector_id
+ * @vf: pointer to the VF info
+ * @vector_id: VF relative vector id
+ *
+ * check for the valid vector id
+ **/
+static inline bool i40e_vc_isvalid_vector_id(struct i40e_vf *vf, u8 vector_id)
+{
+       struct i40e_pf *pf = vf->pf;
+
+       return vector_id < pf->hw.func_caps.num_msix_vectors_vf;
+}
+
+/***********************vf resource mgmt routines*****************/
+
+/**
+ * i40e_vc_get_pf_queue_id
+ * @vf: pointer to the VF info
+ * @vsi_id: id of VSI as provided by the FW
+ * @vsi_queue_id: vsi relative queue id
+ *
+ * return PF relative queue id
+ **/
+static u16 i40e_vc_get_pf_queue_id(struct i40e_vf *vf, u16 vsi_id,
+                                  u8 vsi_queue_id)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = i40e_find_vsi_from_id(pf, vsi_id);
+       u16 pf_queue_id = I40E_QUEUE_END_OF_LIST;
+
+       if (!vsi)
+               return pf_queue_id;
+
+       if (le16_to_cpu(vsi->info.mapping_flags) &
+           I40E_AQ_VSI_QUE_MAP_NONCONTIG)
+               pf_queue_id =
+                       le16_to_cpu(vsi->info.queue_mapping[vsi_queue_id]);
+       else
+               pf_queue_id = le16_to_cpu(vsi->info.queue_mapping[0]) +
+                             vsi_queue_id;
+
+       return pf_queue_id;
+}
+
+/**
+ * i40e_config_irq_link_list
+ * @vf: pointer to the VF info
+ * @vsi_id: id of VSI as given by the FW
+ * @vecmap: irq map info
+ *
+ * configure irq link list from the map
+ **/
+static void i40e_config_irq_link_list(struct i40e_vf *vf, u16 vsi_id,
+                                     struct i40e_virtchnl_vector_map *vecmap)
+{
+       unsigned long linklistmap = 0, tempmap;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       u16 vsi_queue_id, pf_queue_id;
+       enum i40e_queue_type qtype;
+       u16 next_q, vector_id;
+       u32 reg, reg_idx;
+       u16 itr_idx = 0;
+
+       vector_id = vecmap->vector_id;
+       /* setup the head */
+       if (0 == vector_id)
+               reg_idx = I40E_VPINT_LNKLST0(vf->vf_id);
+       else
+               reg_idx = I40E_VPINT_LNKLSTN(
+                    ((pf->hw.func_caps.num_msix_vectors_vf - 1) * vf->vf_id) +
+                    (vector_id - 1));
+
+       if (vecmap->rxq_map == 0 && vecmap->txq_map == 0) {
+               /* Special case - No queues mapped on this vector */
+               wr32(hw, reg_idx, I40E_VPINT_LNKLST0_FIRSTQ_INDX_MASK);
+               goto irq_list_done;
+       }
+       tempmap = vecmap->rxq_map;
+       for_each_set_bit(vsi_queue_id, &tempmap, I40E_MAX_VSI_QP) {
+               linklistmap |= (1 <<
+                               (I40E_VIRTCHNL_SUPPORTED_QTYPES *
+                                vsi_queue_id));
+       }
+
+       tempmap = vecmap->txq_map;
+       for_each_set_bit(vsi_queue_id, &tempmap, I40E_MAX_VSI_QP) {
+               linklistmap |= (1 <<
+                               (I40E_VIRTCHNL_SUPPORTED_QTYPES * vsi_queue_id
+                                + 1));
+       }
+
+       next_q = find_first_bit(&linklistmap,
+                               (I40E_MAX_VSI_QP *
+                                I40E_VIRTCHNL_SUPPORTED_QTYPES));
+       vsi_queue_id = next_q/I40E_VIRTCHNL_SUPPORTED_QTYPES;
+       qtype = next_q%I40E_VIRTCHNL_SUPPORTED_QTYPES;
+       pf_queue_id = i40e_vc_get_pf_queue_id(vf, vsi_id, vsi_queue_id);
+       reg = ((qtype << I40E_VPINT_LNKLSTN_FIRSTQ_TYPE_SHIFT) | pf_queue_id);
+
+       wr32(hw, reg_idx, reg);
+
+       while (next_q < (I40E_MAX_VSI_QP * I40E_VIRTCHNL_SUPPORTED_QTYPES)) {
+               switch (qtype) {
+               case I40E_QUEUE_TYPE_RX:
+                       reg_idx = I40E_QINT_RQCTL(pf_queue_id);
+                       itr_idx = vecmap->rxitr_idx;
+                       break;
+               case I40E_QUEUE_TYPE_TX:
+                       reg_idx = I40E_QINT_TQCTL(pf_queue_id);
+                       itr_idx = vecmap->txitr_idx;
+                       break;
+               default:
+                       break;
+               }
+
+               next_q = find_next_bit(&linklistmap,
+                                      (I40E_MAX_VSI_QP *
+                                       I40E_VIRTCHNL_SUPPORTED_QTYPES),
+                                      next_q + 1);
+               if (next_q <
+                   (I40E_MAX_VSI_QP * I40E_VIRTCHNL_SUPPORTED_QTYPES)) {
+                       vsi_queue_id = next_q / I40E_VIRTCHNL_SUPPORTED_QTYPES;
+                       qtype = next_q % I40E_VIRTCHNL_SUPPORTED_QTYPES;
+                       pf_queue_id = i40e_vc_get_pf_queue_id(vf, vsi_id,
+                                                             vsi_queue_id);
+               } else {
+                       pf_queue_id = I40E_QUEUE_END_OF_LIST;
+                       qtype = 0;
+               }
+
+               /* format for the RQCTL & TQCTL regs is same */
+               reg = (vector_id) |
+                   (qtype << I40E_QINT_RQCTL_NEXTQ_TYPE_SHIFT) |
+                   (pf_queue_id << I40E_QINT_RQCTL_NEXTQ_INDX_SHIFT) |
+                   (1 << I40E_QINT_RQCTL_CAUSE_ENA_SHIFT) |
+                   (itr_idx << I40E_QINT_RQCTL_ITR_INDX_SHIFT);
+               wr32(hw, reg_idx, reg);
+       }
+
+irq_list_done:
+       i40e_flush(hw);
+}
+
+/**
+ * i40e_config_vsi_tx_queue
+ * @vf: pointer to the VF info
+ * @vsi_id: id of VSI as provided by the FW
+ * @vsi_queue_id: vsi relative queue index
+ * @info: config. info
+ *
+ * configure tx queue
+ **/
+static int i40e_config_vsi_tx_queue(struct i40e_vf *vf, u16 vsi_id,
+                                   u16 vsi_queue_id,
+                                   struct i40e_virtchnl_txq_info *info)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_hmc_obj_txq tx_ctx;
+       struct i40e_vsi *vsi;
+       u16 pf_queue_id;
+       u32 qtx_ctl;
+       int ret = 0;
+
+       pf_queue_id = i40e_vc_get_pf_queue_id(vf, vsi_id, vsi_queue_id);
+       vsi = i40e_find_vsi_from_id(pf, vsi_id);
+
+       /* clear the context structure first */
+       memset(&tx_ctx, 0, sizeof(struct i40e_hmc_obj_txq));
+
+       /* only set the required fields */
+       tx_ctx.base = info->dma_ring_addr / 128;
+       tx_ctx.qlen = info->ring_len;
+       tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[0]);
+       tx_ctx.rdylist_act = 0;
+       tx_ctx.head_wb_ena = info->headwb_enabled;
+       tx_ctx.head_wb_addr = info->dma_headwb_addr;
+
+       /* clear the context in the HMC */
+       ret = i40e_clear_lan_tx_queue_context(hw, pf_queue_id);
+       if (ret) {
+               dev_err(&pf->pdev->dev,
+                       "Failed to clear VF LAN Tx queue context %d, error: %d\n",
+                       pf_queue_id, ret);
+               ret = -ENOENT;
+               goto error_context;
+       }
+
+       /* set the context in the HMC */
+       ret = i40e_set_lan_tx_queue_context(hw, pf_queue_id, &tx_ctx);
+       if (ret) {
+               dev_err(&pf->pdev->dev,
+                       "Failed to set VF LAN Tx queue context %d error: %d\n",
+                       pf_queue_id, ret);
+               ret = -ENOENT;
+               goto error_context;
+       }
+
+       /* associate this queue with the PCI VF function */
+       qtx_ctl = I40E_QTX_CTL_VF_QUEUE;
+       qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT)
+                   & I40E_QTX_CTL_PF_INDX_MASK);
+       qtx_ctl |= (((vf->vf_id + hw->func_caps.vf_base_id)
+                    << I40E_QTX_CTL_VFVM_INDX_SHIFT)
+                   & I40E_QTX_CTL_VFVM_INDX_MASK);
+       wr32(hw, I40E_QTX_CTL(pf_queue_id), qtx_ctl);
+       i40e_flush(hw);
+
+error_context:
+       return ret;
+}
+
+/**
+ * i40e_config_vsi_rx_queue
+ * @vf: pointer to the VF info
+ * @vsi_id: id of VSI  as provided by the FW
+ * @vsi_queue_id: vsi relative queue index
+ * @info: config. info
+ *
+ * configure rx queue
+ **/
+static int i40e_config_vsi_rx_queue(struct i40e_vf *vf, u16 vsi_id,
+                                   u16 vsi_queue_id,
+                                   struct i40e_virtchnl_rxq_info *info)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_hmc_obj_rxq rx_ctx;
+       u16 pf_queue_id;
+       int ret = 0;
+
+       pf_queue_id = i40e_vc_get_pf_queue_id(vf, vsi_id, vsi_queue_id);
+
+       /* clear the context structure first */
+       memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
+
+       /* only set the required fields */
+       rx_ctx.base = info->dma_ring_addr / 128;
+       rx_ctx.qlen = info->ring_len;
+
+       if (info->splithdr_enabled) {
+               rx_ctx.hsplit_0 = I40E_RX_SPLIT_L2      |
+                                 I40E_RX_SPLIT_IP      |
+                                 I40E_RX_SPLIT_TCP_UDP |
+                                 I40E_RX_SPLIT_SCTP;
+               /* header length validation */
+               if (info->hdr_size > ((2 * 1024) - 64)) {
+                       ret = -EINVAL;
+                       goto error_param;
+               }
+               rx_ctx.hbuff = info->hdr_size >> I40E_RXQ_CTX_HBUFF_SHIFT;
+
+               /* set splitalways mode 10b */
+               rx_ctx.dtype = 0x2;
+       }
+
+       /* databuffer length validation */
+       if (info->databuffer_size > ((16 * 1024) - 128)) {
+               ret = -EINVAL;
+               goto error_param;
+       }
+       rx_ctx.dbuff = info->databuffer_size >> I40E_RXQ_CTX_DBUFF_SHIFT;
+
+       /* max pkt. length validation */
+       if (info->max_pkt_size >= (16 * 1024) || info->max_pkt_size < 64) {
+               ret = -EINVAL;
+               goto error_param;
+       }
+       rx_ctx.rxmax = info->max_pkt_size;
+
+       /* enable 32bytes desc always */
+       rx_ctx.dsize = 1;
+
+       /* default values */
+       rx_ctx.lrxqthresh = 2;
+       rx_ctx.crcstrip = 1;
+       rx_ctx.prefena = 1;
+       rx_ctx.l2tsel = 1;
+
+       /* clear the context in the HMC */
+       ret = i40e_clear_lan_rx_queue_context(hw, pf_queue_id);
+       if (ret) {
+               dev_err(&pf->pdev->dev,
+                       "Failed to clear VF LAN Rx queue context %d, error: %d\n",
+                       pf_queue_id, ret);
+               ret = -ENOENT;
+               goto error_param;
+       }
+
+       /* set the context in the HMC */
+       ret = i40e_set_lan_rx_queue_context(hw, pf_queue_id, &rx_ctx);
+       if (ret) {
+               dev_err(&pf->pdev->dev,
+                       "Failed to set VF LAN Rx queue context %d error: %d\n",
+                       pf_queue_id, ret);
+               ret = -ENOENT;
+               goto error_param;
+       }
+
+error_param:
+       return ret;
+}
+
+/**
+ * i40e_alloc_vsi_res
+ * @vf: pointer to the VF info
+ * @type: type of VSI to allocate
+ *
+ * alloc VF vsi context & resources
+ **/
+static int i40e_alloc_vsi_res(struct i40e_vf *vf, enum i40e_vsi_type type)
+{
+       struct i40e_mac_filter *f = NULL;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi;
+       int ret = 0;
+
+       vsi = i40e_vsi_setup(pf, type, pf->vsi[pf->lan_vsi]->seid, vf->vf_id);
+
+       if (!vsi) {
+               dev_err(&pf->pdev->dev,
+                       "add vsi failed for VF %d, aq_err %d\n",
+                       vf->vf_id, pf->hw.aq.asq_last_status);
+               ret = -ENOENT;
+               goto error_alloc_vsi_res;
+       }
+       if (type == I40E_VSI_SRIOV) {
+               u8 brdcast[ETH_ALEN] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
+               vf->lan_vsi_idx = vsi->idx;
+               vf->lan_vsi_id = vsi->id;
+               /* If the port VLAN has been configured and then the
+                * VF driver was removed then the VSI port VLAN
+                * configuration was destroyed.  Check if there is
+                * a port VLAN and restore the VSI configuration if
+                * needed.
+                */
+               if (vf->port_vlan_id)
+                       i40e_vsi_add_pvid(vsi, vf->port_vlan_id);
+               f = i40e_add_filter(vsi, vf->default_lan_addr.addr,
+                                   vf->port_vlan_id, true, false);
+               if (!f)
+                       dev_info(&pf->pdev->dev,
+                                "Could not allocate VF MAC addr\n");
+               f = i40e_add_filter(vsi, brdcast, vf->port_vlan_id,
+                                   true, false);
+               if (!f)
+                       dev_info(&pf->pdev->dev,
+                                "Could not allocate VF broadcast filter\n");
+       }
+
+       /* program mac filter */
+       ret = i40e_sync_vsi_filters(vsi);
+       if (ret)
+               dev_err(&pf->pdev->dev, "Unable to program ucast filters\n");
+
+       /* Set VF bandwidth if specified */
+       if (vf->tx_rate) {
+               ret = i40e_aq_config_vsi_bw_limit(&pf->hw, vsi->seid,
+                                                 vf->tx_rate / 50, 0, NULL);
+               if (ret)
+                       dev_err(&pf->pdev->dev, "Unable to set tx rate, VF %d, error code %d.\n",
+                               vf->vf_id, ret);
+       }
+
+error_alloc_vsi_res:
+       return ret;
+}
+
+/**
+ * i40e_enable_vf_mappings
+ * @vf: pointer to the VF info
+ *
+ * enable VF mappings
+ **/
+static void i40e_enable_vf_mappings(struct i40e_vf *vf)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       u32 reg, total_queue_pairs = 0;
+       int j;
+
+       /* Tell the hardware we're using noncontiguous mapping. HW requires
+        * that VF queues be mapped using this method, even when they are
+        * contiguous in real life
+        */
+       wr32(hw, I40E_VSILAN_QBASE(vf->lan_vsi_id),
+            I40E_VSILAN_QBASE_VSIQTABLE_ENA_MASK);
+
+       /* enable VF vplan_qtable mappings */
+       reg = I40E_VPLAN_MAPENA_TXRX_ENA_MASK;
+       wr32(hw, I40E_VPLAN_MAPENA(vf->vf_id), reg);
+
+       /* map PF queues to VF queues */
+       for (j = 0; j < pf->vsi[vf->lan_vsi_idx]->alloc_queue_pairs; j++) {
+               u16 qid = i40e_vc_get_pf_queue_id(vf, vf->lan_vsi_id, j);
+               reg = (qid & I40E_VPLAN_QTABLE_QINDEX_MASK);
+               wr32(hw, I40E_VPLAN_QTABLE(total_queue_pairs, vf->vf_id), reg);
+               total_queue_pairs++;
+       }
+
+       /* map PF queues to VSI */
+       for (j = 0; j < 7; j++) {
+               if (j * 2 >= pf->vsi[vf->lan_vsi_idx]->alloc_queue_pairs) {
+                       reg = 0x07FF07FF;       /* unused */
+               } else {
+                       u16 qid = i40e_vc_get_pf_queue_id(vf, vf->lan_vsi_id,
+                                                         j * 2);
+                       reg = qid;
+                       qid = i40e_vc_get_pf_queue_id(vf, vf->lan_vsi_id,
+                                                     (j * 2) + 1);
+                       reg |= qid << 16;
+               }
+               wr32(hw, I40E_VSILAN_QTABLE(j, vf->lan_vsi_id), reg);
+       }
+
+       i40e_flush(hw);
+}
+
+/**
+ * i40e_disable_vf_mappings
+ * @vf: pointer to the VF info
+ *
+ * disable VF mappings
+ **/
+static void i40e_disable_vf_mappings(struct i40e_vf *vf)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       int i;
+
+       /* disable qp mappings */
+       wr32(hw, I40E_VPLAN_MAPENA(vf->vf_id), 0);
+       for (i = 0; i < I40E_MAX_VSI_QP; i++)
+               wr32(hw, I40E_VPLAN_QTABLE(i, vf->vf_id),
+                    I40E_QUEUE_END_OF_LIST);
+       i40e_flush(hw);
+}
+
+/**
+ * i40e_free_vf_res
+ * @vf: pointer to the VF info
+ *
+ * free VF resources
+ **/
+static void i40e_free_vf_res(struct i40e_vf *vf)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       u32 reg_idx, reg;
+       int i, msix_vf;
+
+       /* free vsi & disconnect it from the parent uplink */
+       if (vf->lan_vsi_idx) {
+               i40e_vsi_release(pf->vsi[vf->lan_vsi_idx]);
+               vf->lan_vsi_idx = 0;
+               vf->lan_vsi_id = 0;
+       }
+       msix_vf = pf->hw.func_caps.num_msix_vectors_vf;
+
+       /* disable interrupts so the VF starts in a known state */
+       for (i = 0; i < msix_vf; i++) {
+               /* format is same for both registers */
+               if (0 == i)
+                       reg_idx = I40E_VFINT_DYN_CTL0(vf->vf_id);
+               else
+                       reg_idx = I40E_VFINT_DYN_CTLN(((msix_vf - 1) *
+                                                     (vf->vf_id))
+                                                    + (i - 1));
+               wr32(hw, reg_idx, I40E_VFINT_DYN_CTLN_CLEARPBA_MASK);
+               i40e_flush(hw);
+       }
+
+       /* clear the irq settings */
+       for (i = 0; i < msix_vf; i++) {
+               /* format is same for both registers */
+               if (0 == i)
+                       reg_idx = I40E_VPINT_LNKLST0(vf->vf_id);
+               else
+                       reg_idx = I40E_VPINT_LNKLSTN(((msix_vf - 1) *
+                                                     (vf->vf_id))
+                                                    + (i - 1));
+               reg = (I40E_VPINT_LNKLSTN_FIRSTQ_TYPE_MASK |
+                      I40E_VPINT_LNKLSTN_FIRSTQ_INDX_MASK);
+               wr32(hw, reg_idx, reg);
+               i40e_flush(hw);
+       }
+       /* reset some of the state varibles keeping
+        * track of the resources
+        */
+       vf->num_queue_pairs = 0;
+       vf->vf_states = 0;
+}
+
+/**
+ * i40e_alloc_vf_res
+ * @vf: pointer to the VF info
+ *
+ * allocate VF resources
+ **/
+static int i40e_alloc_vf_res(struct i40e_vf *vf)
+{
+       struct i40e_pf *pf = vf->pf;
+       int total_queue_pairs = 0;
+       int ret;
+
+       /* allocate hw vsi context & associated resources */
+       ret = i40e_alloc_vsi_res(vf, I40E_VSI_SRIOV);
+       if (ret)
+               goto error_alloc;
+       total_queue_pairs += pf->vsi[vf->lan_vsi_idx]->alloc_queue_pairs;
+       set_bit(I40E_VIRTCHNL_VF_CAP_PRIVILEGE, &vf->vf_caps);
+
+       /* store the total qps number for the runtime
+        * VF req validation
+        */
+       vf->num_queue_pairs = total_queue_pairs;
+
+       /* VF is now completely initialized */
+       set_bit(I40E_VF_STAT_INIT, &vf->vf_states);
+
+error_alloc:
+       if (ret)
+               i40e_free_vf_res(vf);
+
+       return ret;
+}
+
+#define VF_DEVICE_STATUS 0xAA
+#define VF_TRANS_PENDING_MASK 0x20
+/**
+ * i40e_quiesce_vf_pci
+ * @vf: pointer to the VF structure
+ *
+ * Wait for VF PCI transactions to be cleared after reset. Returns -EIO
+ * if the transactions never clear.
+ **/
+static int i40e_quiesce_vf_pci(struct i40e_vf *vf)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       int vf_abs_id, i;
+       u32 reg;
+
+       vf_abs_id = vf->vf_id + hw->func_caps.vf_base_id;
+
+       wr32(hw, I40E_PF_PCI_CIAA,
+            VF_DEVICE_STATUS | (vf_abs_id << I40E_PF_PCI_CIAA_VF_NUM_SHIFT));
+       for (i = 0; i < 100; i++) {
+               reg = rd32(hw, I40E_PF_PCI_CIAD);
+               if ((reg & VF_TRANS_PENDING_MASK) == 0)
+                       return 0;
+               udelay(1);
+       }
+       return -EIO;
+}
+
+/**
+ * i40e_reset_vf
+ * @vf: pointer to the VF structure
+ * @flr: VFLR was issued or not
+ *
+ * reset the VF
+ **/
+void i40e_reset_vf(struct i40e_vf *vf, bool flr)
+{
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       bool rsd = false;
+       int i;
+       u32 reg;
+
+       if (test_and_set_bit(__I40E_VF_DISABLE, &pf->state))
+               return;
+
+       /* warn the VF */
+       clear_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states);
+
+       /* In the case of a VFLR, the HW has already reset the VF and we
+        * just need to clean up, so don't hit the VFRTRIG register.
+        */
+       if (!flr) {
+               /* reset VF using VPGEN_VFRTRIG reg */
+               reg = rd32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id));
+               reg |= I40E_VPGEN_VFRTRIG_VFSWR_MASK;
+               wr32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id), reg);
+               i40e_flush(hw);
+       }
+
+       if (i40e_quiesce_vf_pci(vf))
+               dev_err(&pf->pdev->dev, "VF %d PCI transactions stuck\n",
+                       vf->vf_id);
+
+       /* poll VPGEN_VFRSTAT reg to make sure
+        * that reset is complete
+        */
+       for (i = 0; i < 10; i++) {
+               /* VF reset requires driver to first reset the VF and then
+                * poll the status register to make sure that the reset
+                * completed successfully. Due to internal HW FIFO flushes,
+                * we must wait 10ms before the register will be valid.
+                */
+               usleep_range(10000, 20000);
+               reg = rd32(hw, I40E_VPGEN_VFRSTAT(vf->vf_id));
+               if (reg & I40E_VPGEN_VFRSTAT_VFRD_MASK) {
+                       rsd = true;
+                       break;
+               }
+       }
+
+       if (flr)
+               usleep_range(10000, 20000);
+
+       if (!rsd)
+               dev_err(&pf->pdev->dev, "VF reset check timeout on VF %d\n",
+                       vf->vf_id);
+       wr32(hw, I40E_VFGEN_RSTAT1(vf->vf_id), I40E_VFR_COMPLETED);
+       /* clear the reset bit in the VPGEN_VFRTRIG reg */
+       reg = rd32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id));
+       reg &= ~I40E_VPGEN_VFRTRIG_VFSWR_MASK;
+       wr32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id), reg);
+
+       /* On initial reset, we won't have any queues */
+       if (vf->lan_vsi_idx == 0)
+               goto complete_reset;
+
+       i40e_vsi_control_rings(pf->vsi[vf->lan_vsi_idx], false);
+complete_reset:
+       /* reallocate VF resources to reset the VSI state */
+       i40e_free_vf_res(vf);
+       i40e_alloc_vf_res(vf);
+       i40e_enable_vf_mappings(vf);
+       set_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states);
+
+       /* tell the VF the reset is done */
+       wr32(hw, I40E_VFGEN_RSTAT1(vf->vf_id), I40E_VFR_VFACTIVE);
+       i40e_flush(hw);
+       clear_bit(__I40E_VF_DISABLE, &pf->state);
+}
+
+/**
+ * i40e_free_vfs
+ * @pf: pointer to the PF structure
+ *
+ * free VF resources
+ **/
+void i40e_free_vfs(struct i40e_pf *pf)
+{
+       struct i40e_hw *hw = &pf->hw;
+       u32 reg_idx, bit_idx;
+       int i, tmp, vf_id;
+
+       if (!pf->vf)
+               return;
+       while (test_and_set_bit(__I40E_VF_DISABLE, &pf->state))
+               usleep_range(1000, 2000);
+
+       for (i = 0; i < pf->num_alloc_vfs; i++)
+               if (test_bit(I40E_VF_STAT_INIT, &pf->vf[i].vf_states))
+                       i40e_vsi_control_rings(pf->vsi[pf->vf[i].lan_vsi_idx],
+                                              false);
+
+       /* Disable IOV before freeing resources. This lets any VF drivers
+        * running in the host get themselves cleaned up before we yank
+        * the carpet out from underneath their feet.
+        */
+       if (!pci_vfs_assigned(pf->pdev))
+               pci_disable_sriov(pf->pdev);
+       else
+               dev_warn(&pf->pdev->dev, "VFs are assigned - not disabling SR-IOV\n");
+
+       msleep(20); /* let any messages in transit get finished up */
+
+       /* free up VF resources */
+       tmp = pf->num_alloc_vfs;
+       pf->num_alloc_vfs = 0;
+       for (i = 0; i < tmp; i++) {
+               if (test_bit(I40E_VF_STAT_INIT, &pf->vf[i].vf_states))
+                       i40e_free_vf_res(&pf->vf[i]);
+               /* disable qp mappings */
+               i40e_disable_vf_mappings(&pf->vf[i]);
+       }
+
+       kfree(pf->vf);
+       pf->vf = NULL;
+
+       /* This check is for when the driver is unloaded while VFs are
+        * assigned. Setting the number of VFs to 0 through sysfs is caught
+        * before this function ever gets called.
+        */
+       if (!pci_vfs_assigned(pf->pdev)) {
+               /* Acknowledge VFLR for all VFS. Without this, VFs will fail to
+                * work correctly when SR-IOV gets re-enabled.
+                */
+               for (vf_id = 0; vf_id < tmp; vf_id++) {
+                       reg_idx = (hw->func_caps.vf_base_id + vf_id) / 32;
+                       bit_idx = (hw->func_caps.vf_base_id + vf_id) % 32;
+                       wr32(hw, I40E_GLGEN_VFLRSTAT(reg_idx), (1 << bit_idx));
+               }
+       }
+       clear_bit(__I40E_VF_DISABLE, &pf->state);
+}
+
+#ifdef CONFIG_PCI_IOV
+/**
+ * i40e_alloc_vfs
+ * @pf: pointer to the PF structure
+ * @num_alloc_vfs: number of VFs to allocate
+ *
+ * allocate VF resources
+ **/
+int i40e_alloc_vfs(struct i40e_pf *pf, u16 num_alloc_vfs)
+{
+       struct i40e_vf *vfs;
+       int i, ret = 0;
+
+       /* Disable interrupt 0 so we don't try to handle the VFLR. */
+       i40e_irq_dynamic_disable_icr0(pf);
+
+       /* Check to see if we're just allocating resources for extant VFs */
+       if (pci_num_vf(pf->pdev) != num_alloc_vfs) {
+               ret = pci_enable_sriov(pf->pdev, num_alloc_vfs);
+               if (ret) {
+                       dev_err(&pf->pdev->dev,
+                               "Failed to enable SR-IOV, error %d.\n", ret);
+                       pf->num_alloc_vfs = 0;
+                       goto err_iov;
+               }
+       }
+       /* allocate memory */
+       vfs = kcalloc(num_alloc_vfs, sizeof(struct i40e_vf), GFP_KERNEL);
+       if (!vfs) {
+               ret = -ENOMEM;
+               goto err_alloc;
+       }
+       pf->vf = vfs;
+
+       /* apply default profile */
+       for (i = 0; i < num_alloc_vfs; i++) {
+               vfs[i].pf = pf;
+               vfs[i].parent_type = I40E_SWITCH_ELEMENT_TYPE_VEB;
+               vfs[i].vf_id = i;
+
+               /* assign default capabilities */
+               set_bit(I40E_VIRTCHNL_VF_CAP_L2, &vfs[i].vf_caps);
+               vfs[i].spoofchk = true;
+               /* VF resources get allocated during reset */
+               i40e_reset_vf(&vfs[i], false);
+
+               /* enable VF vplan_qtable mappings */
+               i40e_enable_vf_mappings(&vfs[i]);
+       }
+       pf->num_alloc_vfs = num_alloc_vfs;
+
+err_alloc:
+       if (ret)
+               i40e_free_vfs(pf);
+err_iov:
+       /* Re-enable interrupt 0. */
+       i40e_irq_dynamic_enable_icr0(pf);
+       return ret;
+}
+
+#endif
+/**
+ * i40e_pci_sriov_enable
+ * @pdev: pointer to a pci_dev structure
+ * @num_vfs: number of VFs to allocate
+ *
+ * Enable or change the number of VFs
+ **/
+static int i40e_pci_sriov_enable(struct pci_dev *pdev, int num_vfs)
+{
+#ifdef CONFIG_PCI_IOV
+       struct i40e_pf *pf = pci_get_drvdata(pdev);
+       int pre_existing_vfs = pci_num_vf(pdev);
+       int err = 0;
+
+       dev_info(&pdev->dev, "Allocating %d VFs.\n", num_vfs);
+       if (pre_existing_vfs && pre_existing_vfs != num_vfs)
+               i40e_free_vfs(pf);
+       else if (pre_existing_vfs && pre_existing_vfs == num_vfs)
+               goto out;
+
+       if (num_vfs > pf->num_req_vfs) {
+               err = -EPERM;
+               goto err_out;
+       }
+
+       err = i40e_alloc_vfs(pf, num_vfs);
+       if (err) {
+               dev_warn(&pdev->dev, "Failed to enable SR-IOV: %d\n", err);
+               goto err_out;
+       }
+
+out:
+       return num_vfs;
+
+err_out:
+       return err;
+#endif
+       return 0;
+}
+
+/**
+ * i40e_pci_sriov_configure
+ * @pdev: pointer to a pci_dev structure
+ * @num_vfs: number of VFs to allocate
+ *
+ * Enable or change the number of VFs. Called when the user updates the number
+ * of VFs in sysfs.
+ **/
+int i40e_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
+{
+       struct i40e_pf *pf = pci_get_drvdata(pdev);
+
+       if (num_vfs) {
+               if (!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) {
+                       pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
+                       i40e_do_reset_safe(pf,
+                                          BIT_ULL(__I40E_PF_RESET_REQUESTED));
+               }
+               return i40e_pci_sriov_enable(pdev, num_vfs);
+       }
+
+       if (!pci_vfs_assigned(pf->pdev)) {
+               i40e_free_vfs(pf);
+               pf->flags &= ~I40E_FLAG_VEB_MODE_ENABLED;
+               i40e_do_reset_safe(pf, BIT_ULL(__I40E_PF_RESET_REQUESTED));
+       } else {
+               dev_warn(&pdev->dev, "Unable to free VFs because some are assigned to VMs.\n");
+               return -EINVAL;
+       }
+       return 0;
+}
+
+/***********************virtual channel routines******************/
+
+/**
+ * i40e_vc_send_msg_to_vf
+ * @vf: pointer to the VF info
+ * @v_opcode: virtual channel opcode
+ * @v_retval: virtual channel return value
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * send msg to VF
+ **/
+static int i40e_vc_send_msg_to_vf(struct i40e_vf *vf, u32 v_opcode,
+                                 u32 v_retval, u8 *msg, u16 msglen)
+{
+       struct i40e_pf *pf;
+       struct i40e_hw *hw;
+       int abs_vf_id;
+       i40e_status aq_ret;
+
+       /* validate the request */
+       if (!vf || vf->vf_id >= vf->pf->num_alloc_vfs)
+               return -EINVAL;
+
+       pf = vf->pf;
+       hw = &pf->hw;
+       abs_vf_id = vf->vf_id + hw->func_caps.vf_base_id;
+
+       /* single place to detect unsuccessful return values */
+       if (v_retval) {
+               vf->num_invalid_msgs++;
+               dev_err(&pf->pdev->dev, "Failed opcode %d Error: %d\n",
+                       v_opcode, v_retval);
+               if (vf->num_invalid_msgs >
+                   I40E_DEFAULT_NUM_INVALID_MSGS_ALLOWED) {
+                       dev_err(&pf->pdev->dev,
+                               "Number of invalid messages exceeded for VF %d\n",
+                               vf->vf_id);
+                       dev_err(&pf->pdev->dev, "Use PF Control I/F to enable the VF\n");
+                       set_bit(I40E_VF_STAT_DISABLED, &vf->vf_states);
+               }
+       } else {
+               vf->num_valid_msgs++;
+       }
+
+       aq_ret = i40e_aq_send_msg_to_vf(hw, abs_vf_id,  v_opcode, v_retval,
+                                       msg, msglen, NULL);
+       if (aq_ret) {
+               dev_err(&pf->pdev->dev,
+                       "Unable to send the message to VF %d aq_err %d\n",
+                       vf->vf_id, pf->hw.aq.asq_last_status);
+               return -EIO;
+       }
+
+       return 0;
+}
+
+/**
+ * i40e_vc_send_resp_to_vf
+ * @vf: pointer to the VF info
+ * @opcode: operation code
+ * @retval: return value
+ *
+ * send resp msg to VF
+ **/
+static int i40e_vc_send_resp_to_vf(struct i40e_vf *vf,
+                                  enum i40e_virtchnl_ops opcode,
+                                  i40e_status retval)
+{
+       return i40e_vc_send_msg_to_vf(vf, opcode, retval, NULL, 0);
+}
+
+/**
+ * i40e_vc_get_version_msg
+ * @vf: pointer to the VF info
+ *
+ * called from the VF to request the API version used by the PF
+ **/
+static int i40e_vc_get_version_msg(struct i40e_vf *vf)
+{
+       struct i40e_virtchnl_version_info info = {
+               I40E_VIRTCHNL_VERSION_MAJOR, I40E_VIRTCHNL_VERSION_MINOR
+       };
+
+       return i40e_vc_send_msg_to_vf(vf, I40E_VIRTCHNL_OP_VERSION,
+                                     I40E_SUCCESS, (u8 *)&info,
+                                     sizeof(struct
+                                            i40e_virtchnl_version_info));
+}
+
+/**
+ * i40e_vc_get_vf_resources_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to request its resources
+ **/
+static int i40e_vc_get_vf_resources_msg(struct i40e_vf *vf)
+{
+       struct i40e_virtchnl_vf_resource *vfres = NULL;
+       struct i40e_pf *pf = vf->pf;
+       i40e_status aq_ret = 0;
+       struct i40e_vsi *vsi;
+       int i = 0, len = 0;
+       int num_vsis = 1;
+       int ret;
+
+       if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto err;
+       }
+
+       len = (sizeof(struct i40e_virtchnl_vf_resource) +
+              sizeof(struct i40e_virtchnl_vsi_resource) * num_vsis);
+
+       vfres = kzalloc(len, GFP_KERNEL);
+       if (!vfres) {
+               aq_ret = I40E_ERR_NO_MEMORY;
+               len = 0;
+               goto err;
+       }
+
+       vfres->vf_offload_flags = I40E_VIRTCHNL_VF_OFFLOAD_L2;
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (!vsi->info.pvid)
+               vfres->vf_offload_flags |= I40E_VIRTCHNL_VF_OFFLOAD_VLAN;
+
+       vfres->num_vsis = num_vsis;
+       vfres->num_queue_pairs = vf->num_queue_pairs;
+       vfres->max_vectors = pf->hw.func_caps.num_msix_vectors_vf;
+       if (vf->lan_vsi_idx) {
+               vfres->vsi_res[i].vsi_id = vf->lan_vsi_id;
+               vfres->vsi_res[i].vsi_type = I40E_VSI_SRIOV;
+               vfres->vsi_res[i].num_queue_pairs =
+                   pf->vsi[vf->lan_vsi_idx]->alloc_queue_pairs;
+               memcpy(vfres->vsi_res[i].default_mac_addr,
+                      vf->default_lan_addr.addr, ETH_ALEN);
+               i++;
+       }
+       set_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states);
+
+err:
+       /* send the response back to the VF */
+       ret = i40e_vc_send_msg_to_vf(vf, I40E_VIRTCHNL_OP_GET_VF_RESOURCES,
+                                    aq_ret, (u8 *)vfres, len);
+
+       kfree(vfres);
+       return ret;
+}
+
+/**
+ * i40e_vc_reset_vf_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to reset itself,
+ * unlike other virtchnl messages, PF driver
+ * doesn't send the response back to the VF
+ **/
+static void i40e_vc_reset_vf_msg(struct i40e_vf *vf)
+{
+       if (test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states))
+               i40e_reset_vf(vf, false);
+}
+
+/**
+ * i40e_vc_config_promiscuous_mode_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to configure the promiscuous mode of
+ * VF vsis
+ **/
+static int i40e_vc_config_promiscuous_mode_msg(struct i40e_vf *vf,
+                                              u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_promisc_info *info =
+           (struct i40e_virtchnl_promisc_info *)msg;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_vsi *vsi;
+       bool allmulti = false;
+       i40e_status aq_ret;
+
+       vsi = i40e_find_vsi_from_id(pf, info->vsi_id);
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states) ||
+           !test_bit(I40E_VIRTCHNL_VF_CAP_PRIVILEGE, &vf->vf_caps) ||
+           !i40e_vc_isvalid_vsi_id(vf, info->vsi_id) ||
+           (vsi->type != I40E_VSI_FCOE)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+       if (info->flags & I40E_FLAG_VF_MULTICAST_PROMISC)
+               allmulti = true;
+       aq_ret = i40e_aq_set_vsi_multicast_promiscuous(hw, vsi->seid,
+                                                      allmulti, NULL);
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf,
+                                      I40E_VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE,
+                                      aq_ret);
+}
+
+/**
+ * i40e_vc_config_queues_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to configure the rx/tx
+ * queues
+ **/
+static int i40e_vc_config_queues_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_vsi_queue_config_info *qci =
+           (struct i40e_virtchnl_vsi_queue_config_info *)msg;
+       struct i40e_virtchnl_queue_pair_info *qpi;
+       struct i40e_pf *pf = vf->pf;
+       u16 vsi_id, vsi_queue_id;
+       i40e_status aq_ret = 0;
+       int i;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       vsi_id = qci->vsi_id;
+       if (!i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+       for (i = 0; i < qci->num_queue_pairs; i++) {
+               qpi = &qci->qpair[i];
+               vsi_queue_id = qpi->txq.queue_id;
+               if ((qpi->txq.vsi_id != vsi_id) ||
+                   (qpi->rxq.vsi_id != vsi_id) ||
+                   (qpi->rxq.queue_id != vsi_queue_id) ||
+                   !i40e_vc_isvalid_queue_id(vf, vsi_id, vsi_queue_id)) {
+                       aq_ret = I40E_ERR_PARAM;
+                       goto error_param;
+               }
+
+               if (i40e_config_vsi_rx_queue(vf, vsi_id, vsi_queue_id,
+                                            &qpi->rxq) ||
+                   i40e_config_vsi_tx_queue(vf, vsi_id, vsi_queue_id,
+                                            &qpi->txq)) {
+                       aq_ret = I40E_ERR_PARAM;
+                       goto error_param;
+               }
+       }
+       /* set vsi num_queue_pairs in use to num configured by VF */
+       pf->vsi[vf->lan_vsi_idx]->num_queue_pairs = qci->num_queue_pairs;
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_CONFIG_VSI_QUEUES,
+                                      aq_ret);
+}
+
+/**
+ * i40e_vc_config_irq_map_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to configure the irq to
+ * queue map
+ **/
+static int i40e_vc_config_irq_map_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_irq_map_info *irqmap_info =
+           (struct i40e_virtchnl_irq_map_info *)msg;
+       struct i40e_virtchnl_vector_map *map;
+       u16 vsi_id, vsi_queue_id, vector_id;
+       i40e_status aq_ret = 0;
+       unsigned long tempmap;
+       int i;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       for (i = 0; i < irqmap_info->num_vectors; i++) {
+               map = &irqmap_info->vecmap[i];
+
+               vector_id = map->vector_id;
+               vsi_id = map->vsi_id;
+               /* validate msg params */
+               if (!i40e_vc_isvalid_vector_id(vf, vector_id) ||
+                   !i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+                       aq_ret = I40E_ERR_PARAM;
+                       goto error_param;
+               }
+
+               /* lookout for the invalid queue index */
+               tempmap = map->rxq_map;
+               for_each_set_bit(vsi_queue_id, &tempmap, I40E_MAX_VSI_QP) {
+                       if (!i40e_vc_isvalid_queue_id(vf, vsi_id,
+                                                     vsi_queue_id)) {
+                               aq_ret = I40E_ERR_PARAM;
+                               goto error_param;
+                       }
+               }
+
+               tempmap = map->txq_map;
+               for_each_set_bit(vsi_queue_id, &tempmap, I40E_MAX_VSI_QP) {
+                       if (!i40e_vc_isvalid_queue_id(vf, vsi_id,
+                                                     vsi_queue_id)) {
+                               aq_ret = I40E_ERR_PARAM;
+                               goto error_param;
+                       }
+               }
+
+               i40e_config_irq_link_list(vf, vsi_id, map);
+       }
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_CONFIG_IRQ_MAP,
+                                      aq_ret);
+}
+
+/**
+ * i40e_vc_enable_queues_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to enable all or specific queue(s)
+ **/
+static int i40e_vc_enable_queues_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_queue_select *vqs =
+           (struct i40e_virtchnl_queue_select *)msg;
+       struct i40e_pf *pf = vf->pf;
+       u16 vsi_id = vqs->vsi_id;
+       i40e_status aq_ret = 0;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if (!i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if ((0 == vqs->rx_queues) && (0 == vqs->tx_queues)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if (i40e_vsi_control_rings(pf->vsi[vf->lan_vsi_idx], true))
+               aq_ret = I40E_ERR_TIMEOUT;
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_ENABLE_QUEUES,
+                                      aq_ret);
+}
+
+/**
+ * i40e_vc_disable_queues_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to disable all or specific
+ * queue(s)
+ **/
+static int i40e_vc_disable_queues_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_queue_select *vqs =
+           (struct i40e_virtchnl_queue_select *)msg;
+       struct i40e_pf *pf = vf->pf;
+       i40e_status aq_ret = 0;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if (!i40e_vc_isvalid_vsi_id(vf, vqs->vsi_id)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if ((0 == vqs->rx_queues) && (0 == vqs->tx_queues)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if (i40e_vsi_control_rings(pf->vsi[vf->lan_vsi_idx], false))
+               aq_ret = I40E_ERR_TIMEOUT;
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_DISABLE_QUEUES,
+                                      aq_ret);
+}
+
+/**
+ * i40e_vc_get_stats_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * called from the VF to get vsi stats
+ **/
+static int i40e_vc_get_stats_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_queue_select *vqs =
+           (struct i40e_virtchnl_queue_select *)msg;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_eth_stats stats;
+       i40e_status aq_ret = 0;
+       struct i40e_vsi *vsi;
+
+       memset(&stats, 0, sizeof(struct i40e_eth_stats));
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       if (!i40e_vc_isvalid_vsi_id(vf, vqs->vsi_id)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (!vsi) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+       i40e_update_eth_stats(vsi);
+       stats = vsi->eth_stats;
+
+error_param:
+       /* send the response back to the VF */
+       return i40e_vc_send_msg_to_vf(vf, I40E_VIRTCHNL_OP_GET_STATS, aq_ret,
+                                     (u8 *)&stats, sizeof(stats));
+}
+
+/**
+ * i40e_check_vf_permission
+ * @vf: pointer to the VF info
+ * @macaddr: pointer to the MAC Address being checked
+ *
+ * Check if the VF has permission to add or delete unicast MAC address
+ * filters and return error code -EPERM if not.  Then check if the
+ * address filter requested is broadcast or zero and if so return
+ * an invalid MAC address error code.
+ **/
+static inline int i40e_check_vf_permission(struct i40e_vf *vf, u8 *macaddr)
+{
+       struct i40e_pf *pf = vf->pf;
+       int ret = 0;
+
+       if (is_broadcast_ether_addr(macaddr) ||
+                  is_zero_ether_addr(macaddr)) {
+               dev_err(&pf->pdev->dev, "invalid VF MAC addr %pM\n", macaddr);
+               ret = I40E_ERR_INVALID_MAC_ADDR;
+       } else if (vf->pf_set_mac && !is_multicast_ether_addr(macaddr) &&
+                  !ether_addr_equal(macaddr, vf->default_lan_addr.addr)) {
+               /* If the host VMM administrator has set the VF MAC address
+                * administratively via the ndo_set_vf_mac command then deny
+                * permission to the VF to add or delete unicast MAC addresses.
+                * The VF may request to set the MAC address filter already
+                * assigned to it so do not return an error in that case.
+                */
+               dev_err(&pf->pdev->dev,
+                       "VF attempting to override administratively set MAC address\nPlease reload the VF driver to resume normal operation\n");
+               ret = -EPERM;
+       }
+       return ret;
+}
+
+/**
+ * i40e_vc_add_mac_addr_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * add guest mac address filter
+ **/
+static int i40e_vc_add_mac_addr_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_ether_addr_list *al =
+           (struct i40e_virtchnl_ether_addr_list *)msg;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = NULL;
+       u16 vsi_id = al->vsi_id;
+       i40e_status ret = 0;
+       int i;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states) ||
+           !test_bit(I40E_VIRTCHNL_VF_CAP_PRIVILEGE, &vf->vf_caps) ||
+           !i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+               ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       for (i = 0; i < al->num_elements; i++) {
+               ret = i40e_check_vf_permission(vf, al->list[i].addr);
+               if (ret)
+                       goto error_param;
+       }
+       vsi = pf->vsi[vf->lan_vsi_idx];
+
+       /* add new addresses to the list */
+       for (i = 0; i < al->num_elements; i++) {
+               struct i40e_mac_filter *f;
+
+               f = i40e_find_mac(vsi, al->list[i].addr, true, false);
+               if (!f) {
+                       if (i40e_is_vsi_in_vlan(vsi))
+                               f = i40e_put_mac_in_vlan(vsi, al->list[i].addr,
+                                                        true, false);
+                       else
+                               f = i40e_add_filter(vsi, al->list[i].addr, -1,
+                                                   true, false);
+               }
+
+               if (!f) {
+                       dev_err(&pf->pdev->dev,
+                               "Unable to add VF MAC filter\n");
+                       ret = I40E_ERR_PARAM;
+                       goto error_param;
+               }
+       }
+
+       /* program the updated filter list */
+       if (i40e_sync_vsi_filters(vsi))
+               dev_err(&pf->pdev->dev, "Unable to program VF MAC filters\n");
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_ADD_ETHER_ADDRESS,
+                                      ret);
+}
+
+/**
+ * i40e_vc_del_mac_addr_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * remove guest mac address filter
+ **/
+static int i40e_vc_del_mac_addr_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_ether_addr_list *al =
+           (struct i40e_virtchnl_ether_addr_list *)msg;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = NULL;
+       u16 vsi_id = al->vsi_id;
+       i40e_status ret = 0;
+       int i;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states) ||
+           !test_bit(I40E_VIRTCHNL_VF_CAP_PRIVILEGE, &vf->vf_caps) ||
+           !i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+               ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       for (i = 0; i < al->num_elements; i++) {
+               if (is_broadcast_ether_addr(al->list[i].addr) ||
+                   is_zero_ether_addr(al->list[i].addr)) {
+                       dev_err(&pf->pdev->dev, "invalid VF MAC addr %pM\n",
+                               al->list[i].addr);
+                       ret = I40E_ERR_INVALID_MAC_ADDR;
+                       goto error_param;
+               }
+       }
+       vsi = pf->vsi[vf->lan_vsi_idx];
+
+       /* delete addresses from the list */
+       for (i = 0; i < al->num_elements; i++)
+               i40e_del_filter(vsi, al->list[i].addr,
+                               I40E_VLAN_ANY, true, false);
+
+       /* program the updated filter list */
+       if (i40e_sync_vsi_filters(vsi))
+               dev_err(&pf->pdev->dev, "Unable to program VF MAC filters\n");
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_DEL_ETHER_ADDRESS,
+                                      ret);
+}
+
+/**
+ * i40e_vc_add_vlan_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * program guest vlan id
+ **/
+static int i40e_vc_add_vlan_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_vlan_filter_list *vfl =
+           (struct i40e_virtchnl_vlan_filter_list *)msg;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = NULL;
+       u16 vsi_id = vfl->vsi_id;
+       i40e_status aq_ret = 0;
+       int i;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states) ||
+           !test_bit(I40E_VIRTCHNL_VF_CAP_PRIVILEGE, &vf->vf_caps) ||
+           !i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       for (i = 0; i < vfl->num_elements; i++) {
+               if (vfl->vlan_id[i] > I40E_MAX_VLANID) {
+                       aq_ret = I40E_ERR_PARAM;
+                       dev_err(&pf->pdev->dev,
+                               "invalid VF VLAN id %d\n", vfl->vlan_id[i]);
+                       goto error_param;
+               }
+       }
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (vsi->info.pvid) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       i40e_vlan_stripping_enable(vsi);
+       for (i = 0; i < vfl->num_elements; i++) {
+               /* add new VLAN filter */
+               int ret = i40e_vsi_add_vlan(vsi, vfl->vlan_id[i]);
+               if (ret)
+                       dev_err(&pf->pdev->dev,
+                               "Unable to add VF vlan filter %d, error %d\n",
+                               vfl->vlan_id[i], ret);
+       }
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_ADD_VLAN, aq_ret);
+}
+
+/**
+ * i40e_vc_remove_vlan_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ *
+ * remove programmed guest vlan id
+ **/
+static int i40e_vc_remove_vlan_msg(struct i40e_vf *vf, u8 *msg, u16 msglen)
+{
+       struct i40e_virtchnl_vlan_filter_list *vfl =
+           (struct i40e_virtchnl_vlan_filter_list *)msg;
+       struct i40e_pf *pf = vf->pf;
+       struct i40e_vsi *vsi = NULL;
+       u16 vsi_id = vfl->vsi_id;
+       i40e_status aq_ret = 0;
+       int i;
+
+       if (!test_bit(I40E_VF_STAT_ACTIVE, &vf->vf_states) ||
+           !test_bit(I40E_VIRTCHNL_VF_CAP_PRIVILEGE, &vf->vf_caps) ||
+           !i40e_vc_isvalid_vsi_id(vf, vsi_id)) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       for (i = 0; i < vfl->num_elements; i++) {
+               if (vfl->vlan_id[i] > I40E_MAX_VLANID) {
+                       aq_ret = I40E_ERR_PARAM;
+                       goto error_param;
+               }
+       }
+
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (vsi->info.pvid) {
+               aq_ret = I40E_ERR_PARAM;
+               goto error_param;
+       }
+
+       for (i = 0; i < vfl->num_elements; i++) {
+               int ret = i40e_vsi_kill_vlan(vsi, vfl->vlan_id[i]);
+               if (ret)
+                       dev_err(&pf->pdev->dev,
+                               "Unable to delete VF vlan filter %d, error %d\n",
+                               vfl->vlan_id[i], ret);
+       }
+
+error_param:
+       /* send the response to the VF */
+       return i40e_vc_send_resp_to_vf(vf, I40E_VIRTCHNL_OP_DEL_VLAN, aq_ret);
+}
+
+/**
+ * i40e_vc_validate_vf_msg
+ * @vf: pointer to the VF info
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ * @msghndl: msg handle
+ *
+ * validate msg
+ **/
+static int i40e_vc_validate_vf_msg(struct i40e_vf *vf, u32 v_opcode,
+                                  u32 v_retval, u8 *msg, u16 msglen)
+{
+       bool err_msg_format = false;
+       int valid_len;
+
+       /* Check if VF is disabled. */
+       if (test_bit(I40E_VF_STAT_DISABLED, &vf->vf_states))
+               return I40E_ERR_PARAM;
+
+       /* Validate message length. */
+       switch (v_opcode) {
+       case I40E_VIRTCHNL_OP_VERSION:
+               valid_len = sizeof(struct i40e_virtchnl_version_info);
+               break;
+       case I40E_VIRTCHNL_OP_RESET_VF:
+       case I40E_VIRTCHNL_OP_GET_VF_RESOURCES:
+               valid_len = 0;
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_TX_QUEUE:
+               valid_len = sizeof(struct i40e_virtchnl_txq_info);
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_RX_QUEUE:
+               valid_len = sizeof(struct i40e_virtchnl_rxq_info);
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_VSI_QUEUES:
+               valid_len = sizeof(struct i40e_virtchnl_vsi_queue_config_info);
+               if (msglen >= valid_len) {
+                       struct i40e_virtchnl_vsi_queue_config_info *vqc =
+                           (struct i40e_virtchnl_vsi_queue_config_info *)msg;
+                       valid_len += (vqc->num_queue_pairs *
+                                     sizeof(struct
+                                            i40e_virtchnl_queue_pair_info));
+                       if (vqc->num_queue_pairs == 0)
+                               err_msg_format = true;
+               }
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_IRQ_MAP:
+               valid_len = sizeof(struct i40e_virtchnl_irq_map_info);
+               if (msglen >= valid_len) {
+                       struct i40e_virtchnl_irq_map_info *vimi =
+                           (struct i40e_virtchnl_irq_map_info *)msg;
+                       valid_len += (vimi->num_vectors *
+                                     sizeof(struct i40e_virtchnl_vector_map));
+                       if (vimi->num_vectors == 0)
+                               err_msg_format = true;
+               }
+               break;
+       case I40E_VIRTCHNL_OP_ENABLE_QUEUES:
+       case I40E_VIRTCHNL_OP_DISABLE_QUEUES:
+               valid_len = sizeof(struct i40e_virtchnl_queue_select);
+               break;
+       case I40E_VIRTCHNL_OP_ADD_ETHER_ADDRESS:
+       case I40E_VIRTCHNL_OP_DEL_ETHER_ADDRESS:
+               valid_len = sizeof(struct i40e_virtchnl_ether_addr_list);
+               if (msglen >= valid_len) {
+                       struct i40e_virtchnl_ether_addr_list *veal =
+                           (struct i40e_virtchnl_ether_addr_list *)msg;
+                       valid_len += veal->num_elements *
+                           sizeof(struct i40e_virtchnl_ether_addr);
+                       if (veal->num_elements == 0)
+                               err_msg_format = true;
+               }
+               break;
+       case I40E_VIRTCHNL_OP_ADD_VLAN:
+       case I40E_VIRTCHNL_OP_DEL_VLAN:
+               valid_len = sizeof(struct i40e_virtchnl_vlan_filter_list);
+               if (msglen >= valid_len) {
+                       struct i40e_virtchnl_vlan_filter_list *vfl =
+                           (struct i40e_virtchnl_vlan_filter_list *)msg;
+                       valid_len += vfl->num_elements * sizeof(u16);
+                       if (vfl->num_elements == 0)
+                               err_msg_format = true;
+               }
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE:
+               valid_len = sizeof(struct i40e_virtchnl_promisc_info);
+               break;
+       case I40E_VIRTCHNL_OP_GET_STATS:
+               valid_len = sizeof(struct i40e_virtchnl_queue_select);
+               break;
+       /* These are always errors coming from the VF. */
+       case I40E_VIRTCHNL_OP_EVENT:
+       case I40E_VIRTCHNL_OP_UNKNOWN:
+       default:
+               return -EPERM;
+               break;
+       }
+       /* few more checks */
+       if ((valid_len != msglen) || (err_msg_format)) {
+               i40e_vc_send_resp_to_vf(vf, v_opcode, I40E_ERR_PARAM);
+               return -EINVAL;
+       } else {
+               return 0;
+       }
+}
+
+/**
+ * i40e_vc_process_vf_msg
+ * @pf: pointer to the PF structure
+ * @vf_id: source VF id
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ * @msghndl: msg handle
+ *
+ * called from the common aeq/arq handler to
+ * process request from VF
+ **/
+int i40e_vc_process_vf_msg(struct i40e_pf *pf, u16 vf_id, u32 v_opcode,
+                          u32 v_retval, u8 *msg, u16 msglen)
+{
+       struct i40e_hw *hw = &pf->hw;
+       unsigned int local_vf_id = vf_id - hw->func_caps.vf_base_id;
+       struct i40e_vf *vf;
+       int ret;
+
+       pf->vf_aq_requests++;
+       if (local_vf_id >= pf->num_alloc_vfs)
+               return -EINVAL;
+       vf = &(pf->vf[local_vf_id]);
+       /* perform basic checks on the msg */
+       ret = i40e_vc_validate_vf_msg(vf, v_opcode, v_retval, msg, msglen);
+
+       if (ret) {
+               dev_err(&pf->pdev->dev, "Invalid message from VF %d, opcode %d, len %d\n",
+                       local_vf_id, v_opcode, msglen);
+               return ret;
+       }
+
+       switch (v_opcode) {
+       case I40E_VIRTCHNL_OP_VERSION:
+               ret = i40e_vc_get_version_msg(vf);
+               break;
+       case I40E_VIRTCHNL_OP_GET_VF_RESOURCES:
+               ret = i40e_vc_get_vf_resources_msg(vf);
+               break;
+       case I40E_VIRTCHNL_OP_RESET_VF:
+               i40e_vc_reset_vf_msg(vf);
+               ret = 0;
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE:
+               ret = i40e_vc_config_promiscuous_mode_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_VSI_QUEUES:
+               ret = i40e_vc_config_queues_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_CONFIG_IRQ_MAP:
+               ret = i40e_vc_config_irq_map_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_ENABLE_QUEUES:
+               ret = i40e_vc_enable_queues_msg(vf, msg, msglen);
+               i40e_vc_notify_vf_link_state(vf);
+               break;
+       case I40E_VIRTCHNL_OP_DISABLE_QUEUES:
+               ret = i40e_vc_disable_queues_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_ADD_ETHER_ADDRESS:
+               ret = i40e_vc_add_mac_addr_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_DEL_ETHER_ADDRESS:
+               ret = i40e_vc_del_mac_addr_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_ADD_VLAN:
+               ret = i40e_vc_add_vlan_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_DEL_VLAN:
+               ret = i40e_vc_remove_vlan_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_GET_STATS:
+               ret = i40e_vc_get_stats_msg(vf, msg, msglen);
+               break;
+       case I40E_VIRTCHNL_OP_UNKNOWN:
+       default:
+               dev_err(&pf->pdev->dev, "Unsupported opcode %d from VF %d\n",
+                       v_opcode, local_vf_id);
+               ret = i40e_vc_send_resp_to_vf(vf, v_opcode,
+                                             I40E_ERR_NOT_IMPLEMENTED);
+               break;
+       }
+
+       return ret;
+}
+
+/**
+ * i40e_vc_process_vflr_event
+ * @pf: pointer to the PF structure
+ *
+ * called from the vlfr irq handler to
+ * free up VF resources and state variables
+ **/
+int i40e_vc_process_vflr_event(struct i40e_pf *pf)
+{
+       u32 reg, reg_idx, bit_idx, vf_id;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_vf *vf;
+
+       if (!test_bit(__I40E_VFLR_EVENT_PENDING, &pf->state))
+               return 0;
+
+       /* re-enable vflr interrupt cause */
+       reg = rd32(hw, I40E_PFINT_ICR0_ENA);
+       reg |= I40E_PFINT_ICR0_ENA_VFLR_MASK;
+       wr32(hw, I40E_PFINT_ICR0_ENA, reg);
+       i40e_flush(hw);
+
+       clear_bit(__I40E_VFLR_EVENT_PENDING, &pf->state);
+       for (vf_id = 0; vf_id < pf->num_alloc_vfs; vf_id++) {
+               reg_idx = (hw->func_caps.vf_base_id + vf_id) / 32;
+               bit_idx = (hw->func_caps.vf_base_id + vf_id) % 32;
+               /* read GLGEN_VFLRSTAT register to find out the flr VFs */
+               vf = &pf->vf[vf_id];
+               reg = rd32(hw, I40E_GLGEN_VFLRSTAT(reg_idx));
+               if (reg & (1 << bit_idx)) {
+                       /* clear the bit in GLGEN_VFLRSTAT */
+                       wr32(hw, I40E_GLGEN_VFLRSTAT(reg_idx), (1 << bit_idx));
+
+                       if (!test_bit(__I40E_DOWN, &pf->state))
+                               i40e_reset_vf(vf, true);
+               }
+       }
+
+       return 0;
+}
+
+/**
+ * i40e_ndo_set_vf_mac
+ * @netdev: network interface device structure
+ * @vf_id: VF identifier
+ * @mac: mac address
+ *
+ * program VF mac address
+ **/
+int i40e_ndo_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
+{
+       struct i40e_netdev_priv *np = netdev_priv(netdev);
+       struct i40e_vsi *vsi = np->vsi;
+       struct i40e_pf *pf = vsi->back;
+       struct i40e_mac_filter *f;
+       struct i40e_vf *vf;
+       int ret = 0;
+
+       /* validate the request */
+       if (vf_id >= pf->num_alloc_vfs) {
+               dev_err(&pf->pdev->dev,
+                       "Invalid VF Identifier %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_param;
+       }
+
+       vf = &(pf->vf[vf_id]);
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states)) {
+               dev_err(&pf->pdev->dev,
+                       "Uninitialized VF %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_param;
+       }
+
+       if (!is_valid_ether_addr(mac)) {
+               dev_err(&pf->pdev->dev,
+                       "Invalid VF ethernet address\n");
+               ret = -EINVAL;
+               goto error_param;
+       }
+
+       /* delete the temporary mac address */
+       i40e_del_filter(vsi, vf->default_lan_addr.addr, vf->port_vlan_id,
+                       true, false);
+
+       /* Delete all the filters for this VSI - we're going to kill it
+        * anyway.
+        */
+       list_for_each_entry(f, &vsi->mac_filter_list, list)
+               i40e_del_filter(vsi, f->macaddr, f->vlan, true, false);
+
+       dev_info(&pf->pdev->dev, "Setting MAC %pM on VF %d\n", mac, vf_id);
+       /* program mac filter */
+       if (i40e_sync_vsi_filters(vsi)) {
+               dev_err(&pf->pdev->dev, "Unable to program ucast filters\n");
+               ret = -EIO;
+               goto error_param;
+       }
+       ether_addr_copy(vf->default_lan_addr.addr, mac);
+       vf->pf_set_mac = true;
+       /* Force the VF driver stop so it has to reload with new MAC address */
+       i40e_vc_disable_vf(pf, vf);
+       dev_info(&pf->pdev->dev, "Reload the VF driver to make this change effective.\n");
+
+error_param:
+       return ret;
+}
+
+/**
+ * i40e_ndo_set_vf_port_vlan
+ * @netdev: network interface device structure
+ * @vf_id: VF identifier
+ * @vlan_id: mac address
+ * @qos: priority setting
+ *
+ * program VF vlan id and/or qos
+ **/
+int i40e_ndo_set_vf_port_vlan(struct net_device *netdev,
+                             int vf_id, u16 vlan_id, u8 qos)
+{
+       struct i40e_netdev_priv *np = netdev_priv(netdev);
+       struct i40e_pf *pf = np->vsi->back;
+       struct i40e_vsi *vsi;
+       struct i40e_vf *vf;
+       int ret = 0;
+
+       /* validate the request */
+       if (vf_id >= pf->num_alloc_vfs) {
+               dev_err(&pf->pdev->dev, "Invalid VF Identifier %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_pvid;
+       }
+
+       if ((vlan_id > I40E_MAX_VLANID) || (qos > 7)) {
+               dev_err(&pf->pdev->dev, "Invalid VF Parameters\n");
+               ret = -EINVAL;
+               goto error_pvid;
+       }
+
+       vf = &(pf->vf[vf_id]);
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states)) {
+               dev_err(&pf->pdev->dev, "Uninitialized VF %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_pvid;
+       }
+
+       if (vsi->info.pvid == 0 && i40e_is_vsi_in_vlan(vsi)) {
+               dev_err(&pf->pdev->dev,
+                       "VF %d has already configured VLAN filters and the administrator is requesting a port VLAN override.\nPlease unload and reload the VF driver for this change to take effect.\n",
+                       vf_id);
+               /* Administrator Error - knock the VF offline until he does
+                * the right thing by reconfiguring his network correctly
+                * and then reloading the VF driver.
+                */
+               i40e_vc_disable_vf(pf, vf);
+       }
+
+       /* Check for condition where there was already a port VLAN ID
+        * filter set and now it is being deleted by setting it to zero.
+        * Additionally check for the condition where there was a port
+        * VLAN but now there is a new and different port VLAN being set.
+        * Before deleting all the old VLAN filters we must add new ones
+        * with -1 (I40E_VLAN_ANY) or otherwise we're left with all our
+        * MAC addresses deleted.
+        */
+       if ((!(vlan_id || qos) ||
+           (vlan_id | qos) != le16_to_cpu(vsi->info.pvid)) &&
+           vsi->info.pvid)
+               ret = i40e_vsi_add_vlan(vsi, I40E_VLAN_ANY);
+
+       if (vsi->info.pvid) {
+               /* kill old VLAN */
+               ret = i40e_vsi_kill_vlan(vsi, (le16_to_cpu(vsi->info.pvid) &
+                                              VLAN_VID_MASK));
+               if (ret) {
+                       dev_info(&vsi->back->pdev->dev,
+                                "remove VLAN failed, ret=%d, aq_err=%d\n",
+                                ret, pf->hw.aq.asq_last_status);
+               }
+       }
+       if (vlan_id || qos)
+               ret = i40e_vsi_add_pvid(vsi,
+                               vlan_id | (qos << I40E_VLAN_PRIORITY_SHIFT));
+       else
+               i40e_vsi_remove_pvid(vsi);
+
+       if (vlan_id) {
+               dev_info(&pf->pdev->dev, "Setting VLAN %d, QOS 0x%x on VF %d\n",
+                        vlan_id, qos, vf_id);
+
+               /* add new VLAN filter */
+               ret = i40e_vsi_add_vlan(vsi, vlan_id);
+               if (ret) {
+                       dev_info(&vsi->back->pdev->dev,
+                                "add VF VLAN failed, ret=%d aq_err=%d\n", ret,
+                                vsi->back->hw.aq.asq_last_status);
+                       goto error_pvid;
+               }
+               /* Kill non-vlan MAC filters - ignore error return since
+                * there might not be any non-vlan MAC filters.
+                */
+               i40e_vsi_kill_vlan(vsi, I40E_VLAN_ANY);
+       }
+
+       if (ret) {
+               dev_err(&pf->pdev->dev, "Unable to update VF vsi context\n");
+               goto error_pvid;
+       }
+       /* The Port VLAN needs to be saved across resets the same as the
+        * default LAN MAC address.
+        */
+       vf->port_vlan_id = le16_to_cpu(vsi->info.pvid);
+       ret = 0;
+
+error_pvid:
+       return ret;
+}
+
+#define I40E_BW_CREDIT_DIVISOR 50     /* 50Mbps per BW credit */
+#define I40E_MAX_BW_INACTIVE_ACCUM 4  /* device can accumulate 4 credits max */
+/**
+ * i40e_ndo_set_vf_bw
+ * @netdev: network interface device structure
+ * @vf_id: VF identifier
+ * @tx_rate: Tx rate
+ *
+ * configure VF Tx rate
+ **/
+int i40e_ndo_set_vf_bw(struct net_device *netdev, int vf_id, int min_tx_rate,
+                      int max_tx_rate)
+{
+       struct i40e_netdev_priv *np = netdev_priv(netdev);
+       struct i40e_pf *pf = np->vsi->back;
+       struct i40e_vsi *vsi;
+       struct i40e_vf *vf;
+       int speed = 0;
+       int ret = 0;
+
+       /* validate the request */
+       if (vf_id >= pf->num_alloc_vfs) {
+               dev_err(&pf->pdev->dev, "Invalid VF Identifier %d.\n", vf_id);
+               ret = -EINVAL;
+               goto error;
+       }
+
+       if (min_tx_rate) {
+               dev_err(&pf->pdev->dev, "Invalid min tx rate (%d) (greater than 0) specified for VF %d.\n",
+                       min_tx_rate, vf_id);
+               return -EINVAL;
+       }
+
+       vf = &(pf->vf[vf_id]);
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states)) {
+               dev_err(&pf->pdev->dev, "Uninitialized VF %d.\n", vf_id);
+               ret = -EINVAL;
+               goto error;
+       }
+
+       switch (pf->hw.phy.link_info.link_speed) {
+       case I40E_LINK_SPEED_40GB:
+               speed = 40000;
+               break;
+       case I40E_LINK_SPEED_10GB:
+               speed = 10000;
+               break;
+       case I40E_LINK_SPEED_1GB:
+               speed = 1000;
+               break;
+       default:
+               break;
+       }
+
+       if (max_tx_rate > speed) {
+               dev_err(&pf->pdev->dev, "Invalid max tx rate %d specified for VF %d.",
+                       max_tx_rate, vf->vf_id);
+               ret = -EINVAL;
+               goto error;
+       }
+
+       if ((max_tx_rate < 50) && (max_tx_rate > 0)) {
+               dev_warn(&pf->pdev->dev, "Setting max Tx rate to minimum usable value of 50Mbps.\n");
+               max_tx_rate = 50;
+       }
+
+       /* Tx rate credits are in values of 50Mbps, 0 is disabled*/
+       ret = i40e_aq_config_vsi_bw_limit(&pf->hw, vsi->seid,
+                                         max_tx_rate / I40E_BW_CREDIT_DIVISOR,
+                                         I40E_MAX_BW_INACTIVE_ACCUM, NULL);
+       if (ret) {
+               dev_err(&pf->pdev->dev, "Unable to set max tx rate, error code %d.\n",
+                       ret);
+               ret = -EIO;
+               goto error;
+       }
+       vf->tx_rate = max_tx_rate;
+error:
+       return ret;
+}
+
+/**
+ * i40e_ndo_get_vf_config
+ * @netdev: network interface device structure
+ * @vf_id: VF identifier
+ * @ivi: VF configuration structure
+ *
+ * return VF configuration
+ **/
+int i40e_ndo_get_vf_config(struct net_device *netdev,
+                          int vf_id, struct ifla_vf_info *ivi)
+{
+       struct i40e_netdev_priv *np = netdev_priv(netdev);
+       struct i40e_vsi *vsi = np->vsi;
+       struct i40e_pf *pf = vsi->back;
+       struct i40e_vf *vf;
+       int ret = 0;
+
+       /* validate the request */
+       if (vf_id >= pf->num_alloc_vfs) {
+               dev_err(&pf->pdev->dev, "Invalid VF Identifier %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_param;
+       }
+
+       vf = &(pf->vf[vf_id]);
+       /* first vsi is always the LAN vsi */
+       vsi = pf->vsi[vf->lan_vsi_idx];
+       if (!test_bit(I40E_VF_STAT_INIT, &vf->vf_states)) {
+               dev_err(&pf->pdev->dev, "Uninitialized VF %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_param;
+       }
+
+       ivi->vf = vf_id;
+
+       memcpy(&ivi->mac, vf->default_lan_addr.addr, ETH_ALEN);
+
+       ivi->max_tx_rate = vf->tx_rate;
+       ivi->min_tx_rate = 0;
+       ivi->vlan = le16_to_cpu(vsi->info.pvid) & I40E_VLAN_MASK;
+       ivi->qos = (le16_to_cpu(vsi->info.pvid) & I40E_PRIORITY_MASK) >>
+                  I40E_VLAN_PRIORITY_SHIFT;
+       if (vf->link_forced == false)
+               ivi->linkstate = IFLA_VF_LINK_STATE_AUTO;
+       else if (vf->link_up == true)
+               ivi->linkstate = IFLA_VF_LINK_STATE_ENABLE;
+       else
+               ivi->linkstate = IFLA_VF_LINK_STATE_DISABLE;
+       ivi->spoofchk = vf->spoofchk;
+       ret = 0;
+
+error_param:
+       return ret;
+}
+
+/**
+ * i40e_ndo_set_vf_link_state
+ * @netdev: network interface device structure
+ * @vf_id: VF identifier
+ * @link: required link state
+ *
+ * Set the link state of a specified VF, regardless of physical link state
+ **/
+int i40e_ndo_set_vf_link_state(struct net_device *netdev, int vf_id, int link)
+{
+       struct i40e_netdev_priv *np = netdev_priv(netdev);
+       struct i40e_pf *pf = np->vsi->back;
+       struct i40e_virtchnl_pf_event pfe;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_vf *vf;
+       int abs_vf_id;
+       int ret = 0;
+
+       /* validate the request */
+       if (vf_id >= pf->num_alloc_vfs) {
+               dev_err(&pf->pdev->dev, "Invalid VF Identifier %d\n", vf_id);
+               ret = -EINVAL;
+               goto error_out;
+       }
+
+       vf = &pf->vf[vf_id];
+       abs_vf_id = vf->vf_id + hw->func_caps.vf_base_id;
+
+       pfe.event = I40E_VIRTCHNL_EVENT_LINK_CHANGE;
+       pfe.severity = I40E_PF_EVENT_SEVERITY_INFO;
+
+       switch (link) {
+       case IFLA_VF_LINK_STATE_AUTO:
+               vf->link_forced = false;
+               pfe.event_data.link_event.link_status =
+                       pf->hw.phy.link_info.link_info & I40E_AQ_LINK_UP;
+               pfe.event_data.link_event.link_speed =
+                       pf->hw.phy.link_info.link_speed;
+               break;
+       case IFLA_VF_LINK_STATE_ENABLE:
+               vf->link_forced = true;
+               vf->link_up = true;
+               pfe.event_data.link_event.link_status = true;
+               pfe.event_data.link_event.link_speed = I40E_LINK_SPEED_40GB;
+               break;
+       case IFLA_VF_LINK_STATE_DISABLE:
+               vf->link_forced = true;
+               vf->link_up = false;
+               pfe.event_data.link_event.link_status = false;
+               pfe.event_data.link_event.link_speed = 0;
+               break;
+       default:
+               ret = -EINVAL;
+               goto error_out;
+       }
+       /* Notify the VF of its new link state */
+       i40e_aq_send_msg_to_vf(hw, abs_vf_id, I40E_VIRTCHNL_OP_EVENT,
+                              0, (u8 *)&pfe, sizeof(pfe), NULL);
+
+error_out:
+       return ret;
+}
+
+/**
+ * i40e_ndo_set_vf_spoofchk
+ * @netdev: network interface device structure
+ * @vf_id: VF identifier
+ * @enable: flag to enable or disable feature
+ *
+ * Enable or disable VF spoof checking
+ **/
+int i40e_ndo_set_vf_spoofchk(struct net_device *netdev, int vf_id, bool enable)
+{
+       struct i40e_netdev_priv *np = netdev_priv(netdev);
+       struct i40e_vsi *vsi = np->vsi;
+       struct i40e_pf *pf = vsi->back;
+       struct i40e_vsi_context ctxt;
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_vf *vf;
+       int ret = 0;
+
+       /* validate the request */
+       if (vf_id >= pf->num_alloc_vfs) {
+               dev_err(&pf->pdev->dev, "Invalid VF Identifier %d\n", vf_id);
+               ret = -EINVAL;
+               goto out;
+       }
+
+       vf = &(pf->vf[vf_id]);
+
+       if (enable == vf->spoofchk)
+               goto out;
+
+       vf->spoofchk = enable;
+       memset(&ctxt, 0, sizeof(ctxt));
+       ctxt.seid = pf->vsi[vf->lan_vsi_idx]->seid;
+       ctxt.pf_num = pf->hw.pf_id;
+       ctxt.info.valid_sections = cpu_to_le16(I40E_AQ_VSI_PROP_SECURITY_VALID);
+       if (enable)
+               ctxt.info.sec_flags |= (I40E_AQ_VSI_SEC_FLAG_ENABLE_VLAN_CHK |
+                                       I40E_AQ_VSI_SEC_FLAG_ENABLE_MAC_CHK);
+       ret = i40e_aq_update_vsi_params(hw, &ctxt, NULL);
+       if (ret) {
+               dev_err(&pf->pdev->dev, "Error %d updating VSI parameters\n",
+                       ret);
+               ret = -EIO;
+       }
+out:
+       return ret;
+}