ice: Separate VF VSI initialization/creation from reset flow
authorBrett Creeley <brett.creeley@intel.com>
Sat, 16 May 2020 00:51:10 +0000 (17:51 -0700)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Thu, 28 May 2020 23:01:39 +0000 (16:01 -0700)
Currently the same flow is used for VF VSI initialization/creation and VF
VSI reset. This makes the initialization/creation flow unnecessarily
complicated. Fix this by separating the initialization/creation of the
VF VSI from the reset flow.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/ice/ice_virtchnl_pf.c

index 039f0b0..72a9da3 100644 (file)
@@ -1376,6 +1376,99 @@ static void ice_vc_notify_vf_reset(struct ice_vf *vf)
 }
 
 /**
+ * ice_init_vf_vsi_res - initialize/setup VF VSI resources
+ * @vf: VF to initialize/setup the VSI for
+ *
+ * This function creates a VSI for the VF, adds a VLAN 0 filter, and sets up the
+ * VF VSI's broadcast filter and is only used during initial VF creation.
+ */
+static int ice_init_vf_vsi_res(struct ice_vf *vf)
+{
+       struct ice_pf *pf = vf->pf;
+       u8 broadcast[ETH_ALEN];
+       enum ice_status status;
+       struct ice_vsi *vsi;
+       struct device *dev;
+       int err;
+
+       vf->first_vector_idx = ice_calc_vf_first_vector_idx(pf, vf);
+
+       dev = ice_pf_to_dev(pf);
+       vsi = ice_vf_vsi_setup(pf, pf->hw.port_info, vf->vf_id);
+       if (!vsi) {
+               dev_err(dev, "Failed to create VF VSI\n");
+               return -ENOMEM;
+       }
+
+       vf->lan_vsi_idx = vsi->idx;
+       vf->lan_vsi_num = vsi->vsi_num;
+
+       err = ice_vsi_add_vlan(vsi, 0, ICE_FWD_TO_VSI);
+       if (err) {
+               dev_warn(dev, "Failed to add VLAN 0 filter for VF %d\n",
+                        vf->vf_id);
+               goto release_vsi;
+       }
+
+       eth_broadcast_addr(broadcast);
+       status = ice_fltr_add_mac(vsi, broadcast, ICE_FWD_TO_VSI);
+       if (status) {
+               dev_err(dev, "Failed to add broadcast MAC filter for VF %d, status %s\n",
+                       vf->vf_id, ice_stat_str(status));
+               err = ice_status_to_errno(status);
+               goto release_vsi;
+       }
+
+       vf->num_mac = 1;
+
+       return 0;
+
+release_vsi:
+       ice_vsi_release(vsi);
+       return err;
+}
+
+/**
+ * ice_start_vfs - start VFs so they are ready to be used by SR-IOV
+ * @pf: PF the VFs are associated with
+ */
+static int ice_start_vfs(struct ice_pf *pf)
+{
+       struct ice_hw *hw = &pf->hw;
+       int retval, i;
+
+       ice_for_each_vf(pf, i) {
+               struct ice_vf *vf = &pf->vf[i];
+
+               ice_clear_vf_reset_trigger(vf);
+
+               retval = ice_init_vf_vsi_res(vf);
+               if (retval) {
+                       dev_err(ice_pf_to_dev(pf), "Failed to initialize VSI resources for VF %d, error %d\n",
+                               vf->vf_id, retval);
+                       goto teardown;
+               }
+
+               set_bit(ICE_VF_STATE_INIT, vf->vf_states);
+               ice_ena_vf_mappings(vf);
+               wr32(hw, VFGEN_RSTAT(vf->vf_id), VIRTCHNL_VFR_VFACTIVE);
+       }
+
+       ice_flush(hw);
+       return 0;
+
+teardown:
+       for (i = i - 1; i >= 0; i--) {
+               struct ice_vf *vf = &pf->vf[i];
+
+               ice_dis_vf_mappings(vf);
+               ice_vsi_release(pf->vsi[vf->lan_vsi_idx]);
+       }
+
+       return retval;
+}
+
+/**
  * ice_alloc_vfs - Allocate and set up VFs resources
  * @pf: pointer to the PF structure
  * @num_alloc_vfs: number of VFs to allocate
@@ -1407,6 +1500,13 @@ static int ice_alloc_vfs(struct ice_pf *pf, u16 num_alloc_vfs)
        pf->vf = vfs;
        pf->num_alloc_vfs = num_alloc_vfs;
 
+       if (ice_set_per_vf_res(pf)) {
+               dev_err(dev, "Not enough resources for %d VFs, try with fewer number of VFs\n",
+                       num_alloc_vfs);
+               ret = -ENOSPC;
+               goto err_unroll_sriov;
+       }
+
        /* apply default profile */
        ice_for_each_vf(pf, i) {
                vfs[i].pf = pf;
@@ -1416,15 +1516,17 @@ static int ice_alloc_vfs(struct ice_pf *pf, u16 num_alloc_vfs)
                /* assign default capabilities */
                set_bit(ICE_VIRTCHNL_VF_CAP_L2, &vfs[i].vf_caps);
                vfs[i].spoofchk = true;
+               vfs[i].num_vf_qs = pf->num_qps_per_vf;
        }
 
-       /* VF resources get allocated with initialization */
-       if (!ice_config_res_vfs(pf)) {
-               ret = -EIO;
+       if (ice_start_vfs(pf)) {
+               dev_err(dev, "Failed to start VF(s)\n");
+               ret = -EAGAIN;
                goto err_unroll_sriov;
        }
 
-       return ret;
+       clear_bit(__ICE_VF_DIS, pf->state);
+       return 0;
 
 err_unroll_sriov:
        pf->vf = NULL;