/* * Copyright (c) 2005-2010 Brocade Communications Systems, Inc. * All rights reserved * www.brocade.com * * Linux driver for Brocade Fibre Channel Host Bus Adapter. * * This program is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License (GPL) Version 2 as * published by the Free Software Foundation * * This program is distributed in the hope that 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. */ #include "bfad_drv.h" #include "bfa_fcs.h" #include "bfa_fcbuild.h" #include "bfa_fc.h" BFA_TRC_FILE(FCS, PORT); static void bfa_fcs_lport_send_ls_rjt(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, u8 reason_code, u8 reason_code_expl); static void bfa_fcs_lport_plogi(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, struct fc_logi_s *plogi); static void bfa_fcs_lport_online_actions(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_offline_actions(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_unknown_init(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_unknown_online(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_unknown_offline(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_deleted(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_echo(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, struct fc_echo_s *echo, u16 len); static void bfa_fcs_lport_rnid(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, struct fc_rnid_cmd_s *rnid, u16 len); static void bfa_fs_port_get_gen_topo_data(struct bfa_fcs_lport_s *port, struct fc_rnid_general_topology_data_s *gen_topo_data); static void bfa_fcs_lport_fab_init(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_fab_online(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_fab_offline(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_n2n_init(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_n2n_online(struct bfa_fcs_lport_s *port); static void bfa_fcs_lport_n2n_offline(struct bfa_fcs_lport_s *port); static struct { void (*init) (struct bfa_fcs_lport_s *port); void (*online) (struct bfa_fcs_lport_s *port); void (*offline) (struct bfa_fcs_lport_s *port); } __port_action[] = { { bfa_fcs_lport_unknown_init, bfa_fcs_lport_unknown_online, bfa_fcs_lport_unknown_offline}, { bfa_fcs_lport_fab_init, bfa_fcs_lport_fab_online, bfa_fcs_lport_fab_offline}, { bfa_fcs_lport_n2n_init, bfa_fcs_lport_n2n_online, bfa_fcs_lport_n2n_offline}, }; /* * fcs_port_sm FCS logical port state machine */ enum bfa_fcs_lport_event { BFA_FCS_PORT_SM_CREATE = 1, BFA_FCS_PORT_SM_ONLINE = 2, BFA_FCS_PORT_SM_OFFLINE = 3, BFA_FCS_PORT_SM_DELETE = 4, BFA_FCS_PORT_SM_DELRPORT = 5, BFA_FCS_PORT_SM_STOP = 6, }; static void bfa_fcs_lport_sm_uninit(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event); static void bfa_fcs_lport_sm_init(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event); static void bfa_fcs_lport_sm_online(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event); static void bfa_fcs_lport_sm_offline(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event); static void bfa_fcs_lport_sm_deleting(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event); static void bfa_fcs_lport_sm_stopping(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event); static void bfa_fcs_lport_sm_uninit( struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event) { bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case BFA_FCS_PORT_SM_CREATE: bfa_sm_set_state(port, bfa_fcs_lport_sm_init); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_sm_init(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event) { bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case BFA_FCS_PORT_SM_ONLINE: bfa_sm_set_state(port, bfa_fcs_lport_sm_online); bfa_fcs_lport_online_actions(port); break; case BFA_FCS_PORT_SM_DELETE: bfa_sm_set_state(port, bfa_fcs_lport_sm_uninit); bfa_fcs_lport_deleted(port); break; case BFA_FCS_PORT_SM_STOP: /* If vport - send completion call back */ if (port->vport) bfa_fcs_vport_stop_comp(port->vport); break; case BFA_FCS_PORT_SM_OFFLINE: break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_sm_online( struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event) { struct bfa_fcs_rport_s *rport; struct list_head *qe, *qen; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case BFA_FCS_PORT_SM_OFFLINE: bfa_sm_set_state(port, bfa_fcs_lport_sm_offline); bfa_fcs_lport_offline_actions(port); break; case BFA_FCS_PORT_SM_STOP: __port_action[port->fabric->fab_type].offline(port); if (port->num_rports == 0) { bfa_sm_set_state(port, bfa_fcs_lport_sm_init); /* If vport - send completion call back */ if (port->vport) bfa_fcs_vport_stop_comp(port->vport); } else { bfa_sm_set_state(port, bfa_fcs_lport_sm_stopping); list_for_each_safe(qe, qen, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; bfa_sm_send_event(rport, RPSM_EVENT_DELETE); } } break; case BFA_FCS_PORT_SM_DELETE: __port_action[port->fabric->fab_type].offline(port); if (port->num_rports == 0) { bfa_sm_set_state(port, bfa_fcs_lport_sm_uninit); bfa_fcs_lport_deleted(port); } else { bfa_sm_set_state(port, bfa_fcs_lport_sm_deleting); list_for_each_safe(qe, qen, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; bfa_sm_send_event(rport, RPSM_EVENT_DELETE); } } break; case BFA_FCS_PORT_SM_DELRPORT: break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_sm_offline( struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event) { struct bfa_fcs_rport_s *rport; struct list_head *qe, *qen; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case BFA_FCS_PORT_SM_ONLINE: bfa_sm_set_state(port, bfa_fcs_lport_sm_online); bfa_fcs_lport_online_actions(port); break; case BFA_FCS_PORT_SM_STOP: if (port->num_rports == 0) { bfa_sm_set_state(port, bfa_fcs_lport_sm_init); /* If vport - send completion call back */ if (port->vport) bfa_fcs_vport_stop_comp(port->vport); } else { bfa_sm_set_state(port, bfa_fcs_lport_sm_stopping); list_for_each_safe(qe, qen, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; bfa_sm_send_event(rport, RPSM_EVENT_DELETE); } } break; case BFA_FCS_PORT_SM_DELETE: if (port->num_rports == 0) { bfa_sm_set_state(port, bfa_fcs_lport_sm_uninit); bfa_fcs_lport_deleted(port); } else { bfa_sm_set_state(port, bfa_fcs_lport_sm_deleting); list_for_each_safe(qe, qen, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; bfa_sm_send_event(rport, RPSM_EVENT_DELETE); } } break; case BFA_FCS_PORT_SM_DELRPORT: case BFA_FCS_PORT_SM_OFFLINE: break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_sm_stopping(struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event) { bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case BFA_FCS_PORT_SM_DELRPORT: if (port->num_rports == 0) { bfa_sm_set_state(port, bfa_fcs_lport_sm_init); /* If vport - send completion call back */ if (port->vport) bfa_fcs_vport_stop_comp(port->vport); } break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_sm_deleting( struct bfa_fcs_lport_s *port, enum bfa_fcs_lport_event event) { bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case BFA_FCS_PORT_SM_DELRPORT: if (port->num_rports == 0) { bfa_sm_set_state(port, bfa_fcs_lport_sm_uninit); bfa_fcs_lport_deleted(port); } break; default: bfa_sm_fault(port->fcs, event); } } /* * fcs_port_pvt */ /* * Send a LS reject */ static void bfa_fcs_lport_send_ls_rjt(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, u8 reason_code, u8 reason_code_expl) { struct fchs_s fchs; struct bfa_fcxp_s *fcxp; struct bfa_rport_s *bfa_rport = NULL; int len; bfa_trc(port->fcs, rx_fchs->d_id); bfa_trc(port->fcs, rx_fchs->s_id); fcxp = bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) return; len = fc_ls_rjt_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), rx_fchs->s_id, bfa_fcs_lport_get_fcid(port), rx_fchs->ox_id, reason_code, reason_code_expl); bfa_fcxp_send(fcxp, bfa_rport, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, NULL, NULL, FC_MAX_PDUSZ, 0); } /* * Process incoming plogi from a remote port. */ static void bfa_fcs_lport_plogi(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, struct fc_logi_s *plogi) { struct bfa_fcs_rport_s *rport; bfa_trc(port->fcs, rx_fchs->d_id); bfa_trc(port->fcs, rx_fchs->s_id); /* * If min cfg mode is enabled, drop any incoming PLOGIs */ if (__fcs_min_cfg(port->fcs)) { bfa_trc(port->fcs, rx_fchs->s_id); return; } if (fc_plogi_parse(rx_fchs) != FC_PARSE_OK) { bfa_trc(port->fcs, rx_fchs->s_id); /* * send a LS reject */ bfa_fcs_lport_send_ls_rjt(port, rx_fchs, FC_LS_RJT_RSN_PROTOCOL_ERROR, FC_LS_RJT_EXP_SPARMS_ERR_OPTIONS); return; } /* * Direct Attach P2P mode : verify address assigned by the r-port. */ if ((!bfa_fcs_fabric_is_switched(port->fabric)) && (memcmp((void *)&bfa_fcs_lport_get_pwwn(port), (void *)&plogi->port_name, sizeof(wwn_t)) < 0)) { if (BFA_FCS_PID_IS_WKA(rx_fchs->d_id)) { /* Address assigned to us cannot be a WKA */ bfa_fcs_lport_send_ls_rjt(port, rx_fchs, FC_LS_RJT_RSN_PROTOCOL_ERROR, FC_LS_RJT_EXP_INVALID_NPORT_ID); return; } port->pid = rx_fchs->d_id; bfa_lps_set_n2n_pid(port->fabric->lps, rx_fchs->d_id); } /* * First, check if we know the device by pwwn. */ rport = bfa_fcs_lport_get_rport_by_pwwn(port, plogi->port_name); if (rport) { /* * Direct Attach P2P mode : handle address assigned by r-port. */ if ((!bfa_fcs_fabric_is_switched(port->fabric)) && (memcmp((void *)&bfa_fcs_lport_get_pwwn(port), (void *)&plogi->port_name, sizeof(wwn_t)) < 0)) { port->pid = rx_fchs->d_id; bfa_lps_set_n2n_pid(port->fabric->lps, rx_fchs->d_id); rport->pid = rx_fchs->s_id; } bfa_fcs_rport_plogi(rport, rx_fchs, plogi); return; } /* * Next, lookup rport by PID. */ rport = bfa_fcs_lport_get_rport_by_pid(port, rx_fchs->s_id); if (!rport) { /* * Inbound PLOGI from a new device. */ bfa_fcs_rport_plogi_create(port, rx_fchs, plogi); return; } /* * Rport is known only by PID. */ if (rport->pwwn) { /* * This is a different device with the same pid. Old device * disappeared. Send implicit LOGO to old device. */ WARN_ON(rport->pwwn == plogi->port_name); bfa_sm_send_event(rport, RPSM_EVENT_LOGO_IMP); /* * Inbound PLOGI from a new device (with old PID). */ bfa_fcs_rport_plogi_create(port, rx_fchs, plogi); return; } /* * PLOGI crossing each other. */ WARN_ON(rport->pwwn != WWN_NULL); bfa_fcs_rport_plogi(rport, rx_fchs, plogi); } /* * Process incoming ECHO. * Since it does not require a login, it is processed here. */ static void bfa_fcs_lport_echo(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, struct fc_echo_s *echo, u16 rx_len) { struct fchs_s fchs; struct bfa_fcxp_s *fcxp; struct bfa_rport_s *bfa_rport = NULL; int len, pyld_len; bfa_trc(port->fcs, rx_fchs->s_id); bfa_trc(port->fcs, rx_fchs->d_id); fcxp = bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) return; len = fc_ls_acc_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), rx_fchs->s_id, bfa_fcs_lport_get_fcid(port), rx_fchs->ox_id); /* * Copy the payload (if any) from the echo frame */ pyld_len = rx_len - sizeof(struct fchs_s); bfa_trc(port->fcs, rx_len); bfa_trc(port->fcs, pyld_len); if (pyld_len > len) memcpy(((u8 *) bfa_fcxp_get_reqbuf(fcxp)) + sizeof(struct fc_echo_s), (echo + 1), (pyld_len - sizeof(struct fc_echo_s))); bfa_fcxp_send(fcxp, bfa_rport, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, pyld_len, &fchs, NULL, NULL, FC_MAX_PDUSZ, 0); } /* * Process incoming RNID. * Since it does not require a login, it is processed here. */ static void bfa_fcs_lport_rnid(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs, struct fc_rnid_cmd_s *rnid, u16 rx_len) { struct fc_rnid_common_id_data_s common_id_data; struct fc_rnid_general_topology_data_s gen_topo_data; struct fchs_s fchs; struct bfa_fcxp_s *fcxp; struct bfa_rport_s *bfa_rport = NULL; u16 len; u32 data_format; bfa_trc(port->fcs, rx_fchs->s_id); bfa_trc(port->fcs, rx_fchs->d_id); bfa_trc(port->fcs, rx_len); fcxp = bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) return; /* * Check Node Indentification Data Format * We only support General Topology Discovery Format. * For any other requested Data Formats, we return Common Node Id Data * only, as per FC-LS. */ bfa_trc(port->fcs, rnid->node_id_data_format); if (rnid->node_id_data_format == RNID_NODEID_DATA_FORMAT_DISCOVERY) { data_format = RNID_NODEID_DATA_FORMAT_DISCOVERY; /* * Get General topology data for this port */ bfa_fs_port_get_gen_topo_data(port, &gen_topo_data); } else { data_format = RNID_NODEID_DATA_FORMAT_COMMON; } /* * Copy the Node Id Info */ common_id_data.port_name = bfa_fcs_lport_get_pwwn(port); common_id_data.node_name = bfa_fcs_lport_get_nwwn(port); len = fc_rnid_acc_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), rx_fchs->s_id, bfa_fcs_lport_get_fcid(port), rx_fchs->ox_id, data_format, &common_id_data, &gen_topo_data); bfa_fcxp_send(fcxp, bfa_rport, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, NULL, NULL, FC_MAX_PDUSZ, 0); } /* * Fill out General Topolpgy Discovery Data for RNID ELS. */ static void bfa_fs_port_get_gen_topo_data(struct bfa_fcs_lport_s *port, struct fc_rnid_general_topology_data_s *gen_topo_data) { memset(gen_topo_data, 0, sizeof(struct fc_rnid_general_topology_data_s)); gen_topo_data->asso_type = cpu_to_be32(RNID_ASSOCIATED_TYPE_HOST); gen_topo_data->phy_port_num = 0; /* @todo */ gen_topo_data->num_attached_nodes = cpu_to_be32(1); } static void bfa_fcs_lport_online_actions(struct bfa_fcs_lport_s *port) { struct bfad_s *bfad = (struct bfad_s *)port->fcs->bfad; char lpwwn_buf[BFA_STRING_32]; bfa_trc(port->fcs, port->fabric->oper_type); __port_action[port->fabric->fab_type].init(port); __port_action[port->fabric->fab_type].online(port); wwn2str(lpwwn_buf, bfa_fcs_lport_get_pwwn(port)); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Logical port online: WWN = %s Role = %s\n", lpwwn_buf, "Initiator"); bfad->bfad_flags |= BFAD_PORT_ONLINE; } static void bfa_fcs_lport_offline_actions(struct bfa_fcs_lport_s *port) { struct list_head *qe, *qen; struct bfa_fcs_rport_s *rport; struct bfad_s *bfad = (struct bfad_s *)port->fcs->bfad; char lpwwn_buf[BFA_STRING_32]; bfa_trc(port->fcs, port->fabric->oper_type); __port_action[port->fabric->fab_type].offline(port); wwn2str(lpwwn_buf, bfa_fcs_lport_get_pwwn(port)); if (bfa_sm_cmp_state(port->fabric, bfa_fcs_fabric_sm_online) == BFA_TRUE) BFA_LOG(KERN_ERR, bfad, bfa_log_level, "Logical port lost fabric connectivity: WWN = %s Role = %s\n", lpwwn_buf, "Initiator"); else BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Logical port taken offline: WWN = %s Role = %s\n", lpwwn_buf, "Initiator"); list_for_each_safe(qe, qen, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; bfa_sm_send_event(rport, RPSM_EVENT_LOGO_IMP); } } static void bfa_fcs_lport_unknown_init(struct bfa_fcs_lport_s *port) { WARN_ON(1); } static void bfa_fcs_lport_unknown_online(struct bfa_fcs_lport_s *port) { WARN_ON(1); } static void bfa_fcs_lport_unknown_offline(struct bfa_fcs_lport_s *port) { WARN_ON(1); } static void bfa_fcs_lport_abts_acc(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs) { struct fchs_s fchs; struct bfa_fcxp_s *fcxp; int len; bfa_trc(port->fcs, rx_fchs->d_id); bfa_trc(port->fcs, rx_fchs->s_id); fcxp = bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) return; len = fc_ba_acc_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), rx_fchs->s_id, bfa_fcs_lport_get_fcid(port), rx_fchs->ox_id, 0); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, NULL, NULL, FC_MAX_PDUSZ, 0); } static void bfa_fcs_lport_deleted(struct bfa_fcs_lport_s *port) { struct bfad_s *bfad = (struct bfad_s *)port->fcs->bfad; char lpwwn_buf[BFA_STRING_32]; wwn2str(lpwwn_buf, bfa_fcs_lport_get_pwwn(port)); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Logical port deleted: WWN = %s Role = %s\n", lpwwn_buf, "Initiator"); /* Base port will be deleted by the OS driver */ if (port->vport) { bfa_fcb_lport_delete(port->fcs->bfad, port->port_cfg.roles, port->fabric->vf_drv, port->vport ? port->vport->vport_drv : NULL); bfa_fcs_vport_delete_comp(port->vport); } else { bfa_wc_down(&port->fabric->wc); } } /* * Unsolicited frame receive handling. */ void bfa_fcs_lport_uf_recv(struct bfa_fcs_lport_s *lport, struct fchs_s *fchs, u16 len) { u32 pid = fchs->s_id; struct bfa_fcs_rport_s *rport = NULL; struct fc_els_cmd_s *els_cmd = (struct fc_els_cmd_s *) (fchs + 1); bfa_stats(lport, uf_recvs); bfa_trc(lport->fcs, fchs->type); if (!bfa_fcs_lport_is_online(lport)) { bfa_stats(lport, uf_recv_drops); return; } /* * First, handle ELSs that donot require a login. */ /* * Handle PLOGI first */ if ((fchs->type == FC_TYPE_ELS) && (els_cmd->els_code == FC_ELS_PLOGI)) { bfa_fcs_lport_plogi(lport, fchs, (struct fc_logi_s *) els_cmd); return; } /* * Handle ECHO separately. */ if ((fchs->type == FC_TYPE_ELS) && (els_cmd->els_code == FC_ELS_ECHO)) { bfa_fcs_lport_echo(lport, fchs, (struct fc_echo_s *)els_cmd, len); return; } /* * Handle RNID separately. */ if ((fchs->type == FC_TYPE_ELS) && (els_cmd->els_code == FC_ELS_RNID)) { bfa_fcs_lport_rnid(lport, fchs, (struct fc_rnid_cmd_s *) els_cmd, len); return; } if (fchs->type == FC_TYPE_BLS) { if ((fchs->routing == FC_RTG_BASIC_LINK) && (fchs->cat_info == FC_CAT_ABTS)) bfa_fcs_lport_abts_acc(lport, fchs); return; } /* * look for a matching remote port ID */ rport = bfa_fcs_lport_get_rport_by_pid(lport, pid); if (rport) { bfa_trc(rport->fcs, fchs->s_id); bfa_trc(rport->fcs, fchs->d_id); bfa_trc(rport->fcs, fchs->type); bfa_fcs_rport_uf_recv(rport, fchs, len); return; } /* * Only handles ELS frames for now. */ if (fchs->type != FC_TYPE_ELS) { bfa_trc(lport->fcs, fchs->s_id); bfa_trc(lport->fcs, fchs->d_id); /* ignore type FC_TYPE_FC_FSS */ if (fchs->type != FC_TYPE_FC_FSS) bfa_sm_fault(lport->fcs, fchs->type); return; } bfa_trc(lport->fcs, els_cmd->els_code); if (els_cmd->els_code == FC_ELS_RSCN) { bfa_fcs_lport_scn_process_rscn(lport, fchs, len); return; } if (els_cmd->els_code == FC_ELS_LOGO) { /* * @todo Handle LOGO frames received. */ return; } if (els_cmd->els_code == FC_ELS_PRLI) { /* * @todo Handle PRLI frames received. */ return; } /* * Unhandled ELS frames. Send a LS_RJT. */ bfa_fcs_lport_send_ls_rjt(lport, fchs, FC_LS_RJT_RSN_CMD_NOT_SUPP, FC_LS_RJT_EXP_NO_ADDL_INFO); } /* * PID based Lookup for a R-Port in the Port R-Port Queue */ struct bfa_fcs_rport_s * bfa_fcs_lport_get_rport_by_pid(struct bfa_fcs_lport_s *port, u32 pid) { struct bfa_fcs_rport_s *rport; struct list_head *qe; list_for_each(qe, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; if (rport->pid == pid) return rport; } bfa_trc(port->fcs, pid); return NULL; } /* * PWWN based Lookup for a R-Port in the Port R-Port Queue */ struct bfa_fcs_rport_s * bfa_fcs_lport_get_rport_by_pwwn(struct bfa_fcs_lport_s *port, wwn_t pwwn) { struct bfa_fcs_rport_s *rport; struct list_head *qe; list_for_each(qe, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; if (wwn_is_equal(rport->pwwn, pwwn)) return rport; } bfa_trc(port->fcs, pwwn); return NULL; } /* * NWWN based Lookup for a R-Port in the Port R-Port Queue */ struct bfa_fcs_rport_s * bfa_fcs_lport_get_rport_by_nwwn(struct bfa_fcs_lport_s *port, wwn_t nwwn) { struct bfa_fcs_rport_s *rport; struct list_head *qe; list_for_each(qe, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; if (wwn_is_equal(rport->nwwn, nwwn)) return rport; } bfa_trc(port->fcs, nwwn); return NULL; } /* * Called by rport module when new rports are discovered. */ void bfa_fcs_lport_add_rport( struct bfa_fcs_lport_s *port, struct bfa_fcs_rport_s *rport) { list_add_tail(&rport->qe, &port->rport_q); port->num_rports++; } /* * Called by rport module to when rports are deleted. */ void bfa_fcs_lport_del_rport( struct bfa_fcs_lport_s *port, struct bfa_fcs_rport_s *rport) { WARN_ON(!bfa_q_is_on_q(&port->rport_q, rport)); list_del(&rport->qe); port->num_rports--; bfa_sm_send_event(port, BFA_FCS_PORT_SM_DELRPORT); } /* * Called by fabric for base port when fabric login is complete. * Called by vport for virtual ports when FDISC is complete. */ void bfa_fcs_lport_online(struct bfa_fcs_lport_s *port) { bfa_sm_send_event(port, BFA_FCS_PORT_SM_ONLINE); } /* * Called by fabric for base port when fabric goes offline. * Called by vport for virtual ports when virtual port becomes offline. */ void bfa_fcs_lport_offline(struct bfa_fcs_lport_s *port) { bfa_sm_send_event(port, BFA_FCS_PORT_SM_OFFLINE); } /* * Called by fabric to delete base lport and associated resources. * * Called by vport to delete lport and associated resources. Should call * bfa_fcs_vport_delete_comp() for vports on completion. */ void bfa_fcs_lport_delete(struct bfa_fcs_lport_s *port) { bfa_sm_send_event(port, BFA_FCS_PORT_SM_DELETE); } /* * Return TRUE if port is online, else return FALSE */ bfa_boolean_t bfa_fcs_lport_is_online(struct bfa_fcs_lport_s *port) { return bfa_sm_cmp_state(port, bfa_fcs_lport_sm_online); } /* * Attach time initialization of logical ports. */ void bfa_fcs_lport_attach(struct bfa_fcs_lport_s *lport, struct bfa_fcs_s *fcs, u16 vf_id, struct bfa_fcs_vport_s *vport) { lport->fcs = fcs; lport->fabric = bfa_fcs_vf_lookup(fcs, vf_id); lport->vport = vport; lport->lp_tag = (vport) ? vport->lps->lp_tag : lport->fabric->lps->lp_tag; INIT_LIST_HEAD(&lport->rport_q); lport->num_rports = 0; } /* * Logical port initialization of base or virtual port. * Called by fabric for base port or by vport for virtual ports. */ void bfa_fcs_lport_init(struct bfa_fcs_lport_s *lport, struct bfa_lport_cfg_s *port_cfg) { struct bfa_fcs_vport_s *vport = lport->vport; struct bfad_s *bfad = (struct bfad_s *)lport->fcs->bfad; char lpwwn_buf[BFA_STRING_32]; lport->port_cfg = *port_cfg; lport->bfad_port = bfa_fcb_lport_new(lport->fcs->bfad, lport, lport->port_cfg.roles, lport->fabric->vf_drv, vport ? vport->vport_drv : NULL); wwn2str(lpwwn_buf, bfa_fcs_lport_get_pwwn(lport)); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "New logical port created: WWN = %s Role = %s\n", lpwwn_buf, "Initiator"); bfa_sm_set_state(lport, bfa_fcs_lport_sm_uninit); bfa_sm_send_event(lport, BFA_FCS_PORT_SM_CREATE); } /* * fcs_lport_api */ void bfa_fcs_lport_get_attr( struct bfa_fcs_lport_s *port, struct bfa_lport_attr_s *port_attr) { if (bfa_sm_cmp_state(port, bfa_fcs_lport_sm_online)) port_attr->pid = port->pid; else port_attr->pid = 0; port_attr->port_cfg = port->port_cfg; if (port->fabric) { port_attr->port_type = port->fabric->oper_type; port_attr->loopback = bfa_sm_cmp_state(port->fabric, bfa_fcs_fabric_sm_loopback); port_attr->authfail = bfa_sm_cmp_state(port->fabric, bfa_fcs_fabric_sm_auth_failed); port_attr->fabric_name = bfa_fcs_lport_get_fabric_name(port); memcpy(port_attr->fabric_ip_addr, bfa_fcs_lport_get_fabric_ipaddr(port), BFA_FCS_FABRIC_IPADDR_SZ); if (port->vport != NULL) { port_attr->port_type = BFA_PORT_TYPE_VPORT; port_attr->fpma_mac = port->vport->lps->lp_mac; } else { port_attr->fpma_mac = port->fabric->lps->lp_mac; } } else { port_attr->port_type = BFA_PORT_TYPE_UNKNOWN; port_attr->state = BFA_LPORT_UNINIT; } } /* * bfa_fcs_lport_fab port fab functions */ /* * Called by port to initialize fabric services of the base port. */ static void bfa_fcs_lport_fab_init(struct bfa_fcs_lport_s *port) { bfa_fcs_lport_ns_init(port); bfa_fcs_lport_scn_init(port); bfa_fcs_lport_ms_init(port); } /* * Called by port to notify transition to online state. */ static void bfa_fcs_lport_fab_online(struct bfa_fcs_lport_s *port) { bfa_fcs_lport_ns_online(port); bfa_fcs_lport_scn_online(port); } /* * Called by port to notify transition to offline state. */ static void bfa_fcs_lport_fab_offline(struct bfa_fcs_lport_s *port) { bfa_fcs_lport_ns_offline(port); bfa_fcs_lport_scn_offline(port); bfa_fcs_lport_ms_offline(port); } /* * bfa_fcs_lport_n2n functions */ /* * Called by fcs/port to initialize N2N topology. */ static void bfa_fcs_lport_n2n_init(struct bfa_fcs_lport_s *port) { } /* * Called by fcs/port to notify transition to online state. */ static void bfa_fcs_lport_n2n_online(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_n2n_s *n2n_port = &port->port_topo.pn2n; struct bfa_lport_cfg_s *pcfg = &port->port_cfg; struct bfa_fcs_rport_s *rport; bfa_trc(port->fcs, pcfg->pwwn); /* * If our PWWN is > than that of the r-port, we have to initiate PLOGI * and assign an Address. if not, we need to wait for its PLOGI. * * If our PWWN is < than that of the remote port, it will send a PLOGI * with the PIDs assigned. The rport state machine take care of this * incoming PLOGI. */ if (memcmp ((void *)&pcfg->pwwn, (void *)&n2n_port->rem_port_wwn, sizeof(wwn_t)) > 0) { port->pid = N2N_LOCAL_PID; bfa_lps_set_n2n_pid(port->fabric->lps, N2N_LOCAL_PID); /* * First, check if we know the device by pwwn. */ rport = bfa_fcs_lport_get_rport_by_pwwn(port, n2n_port->rem_port_wwn); if (rport) { bfa_trc(port->fcs, rport->pid); bfa_trc(port->fcs, rport->pwwn); rport->pid = N2N_REMOTE_PID; bfa_sm_send_event(rport, RPSM_EVENT_PLOGI_SEND); return; } /* * In n2n there can be only one rport. Delete the old one * whose pid should be zero, because it is offline. */ if (port->num_rports > 0) { rport = bfa_fcs_lport_get_rport_by_pid(port, 0); WARN_ON(rport == NULL); if (rport) { bfa_trc(port->fcs, rport->pwwn); bfa_sm_send_event(rport, RPSM_EVENT_DELETE); } } bfa_fcs_rport_create(port, N2N_REMOTE_PID); } } /* * Called by fcs/port to notify transition to offline state. */ static void bfa_fcs_lport_n2n_offline(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_n2n_s *n2n_port = &port->port_topo.pn2n; bfa_trc(port->fcs, port->pid); port->pid = 0; n2n_port->rem_port_wwn = 0; n2n_port->reply_oxid = 0; } #define BFA_FCS_FDMI_CMD_MAX_RETRIES 2 /* * forward declarations */ static void bfa_fcs_lport_fdmi_send_rhba(void *fdmi_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_fdmi_send_rprt(void *fdmi_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_fdmi_send_rpa(void *fdmi_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_fdmi_rhba_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_fdmi_rprt_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_fdmi_rpa_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_fdmi_timeout(void *arg); static u16 bfa_fcs_lport_fdmi_build_rhba_pyld(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld); static u16 bfa_fcs_lport_fdmi_build_rprt_pyld(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld); static u16 bfa_fcs_lport_fdmi_build_rpa_pyld(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld); static u16 bfa_fcs_lport_fdmi_build_portattr_block(struct bfa_fcs_lport_fdmi_s * fdmi, u8 *pyld); static void bfa_fcs_fdmi_get_hbaattr(struct bfa_fcs_lport_fdmi_s *fdmi, struct bfa_fcs_fdmi_hba_attr_s *hba_attr); static void bfa_fcs_fdmi_get_portattr(struct bfa_fcs_lport_fdmi_s *fdmi, struct bfa_fcs_fdmi_port_attr_s *port_attr); /* * fcs_fdmi_sm FCS FDMI state machine */ /* * FDMI State Machine events */ enum port_fdmi_event { FDMISM_EVENT_PORT_ONLINE = 1, FDMISM_EVENT_PORT_OFFLINE = 2, FDMISM_EVENT_RSP_OK = 4, FDMISM_EVENT_RSP_ERROR = 5, FDMISM_EVENT_TIMEOUT = 6, FDMISM_EVENT_RHBA_SENT = 7, FDMISM_EVENT_RPRT_SENT = 8, FDMISM_EVENT_RPA_SENT = 9, }; static void bfa_fcs_lport_fdmi_sm_offline(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_sending_rhba( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_rhba(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_rhba_retry( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_sending_rprt( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_rprt(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_rprt_retry( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_sending_rpa( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_rpa(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_rpa_retry( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_online(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); static void bfa_fcs_lport_fdmi_sm_disabled( struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event); /* * Start in offline state - awaiting MS to send start. */ static void bfa_fcs_lport_fdmi_sm_offline(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); fdmi->retry_cnt = 0; switch (event) { case FDMISM_EVENT_PORT_ONLINE: if (port->vport) { /* * For Vports, register a new port. */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_sending_rprt); bfa_fcs_lport_fdmi_send_rprt(fdmi, NULL); } else { /* * For a base port, we should first register the HBA * attribute. The HBA attribute also contains the base * port registration. */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_sending_rhba); bfa_fcs_lport_fdmi_send_rhba(fdmi, NULL); } break; case FDMISM_EVENT_PORT_OFFLINE: break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_sending_rhba(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_RHBA_SENT: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_rhba); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(port), &fdmi->fcxp_wqe); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_rhba(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_RSP_ERROR: /* * if max retries have not been reached, start timer for a * delayed retry */ if (fdmi->retry_cnt++ < BFA_FCS_FDMI_CMD_MAX_RETRIES) { bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_rhba_retry); bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(port), &fdmi->timer, bfa_fcs_lport_fdmi_timeout, fdmi, BFA_FCS_RETRY_TIMEOUT); } else { /* * set state to offline */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); } break; case FDMISM_EVENT_RSP_OK: /* * Initiate Register Port Attributes */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_sending_rpa); fdmi->retry_cnt = 0; bfa_fcs_lport_fdmi_send_rpa(fdmi, NULL); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_fcxp_discard(fdmi->fcxp); bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_rhba_retry(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_sending_rhba); bfa_fcs_lport_fdmi_send_rhba(fdmi, NULL); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); bfa_timer_stop(&fdmi->timer); break; default: bfa_sm_fault(port->fcs, event); } } /* * RPRT : Register Port */ static void bfa_fcs_lport_fdmi_sm_sending_rprt(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_RPRT_SENT: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_rprt); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(port), &fdmi->fcxp_wqe); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_rprt(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_RSP_ERROR: /* * if max retries have not been reached, start timer for a * delayed retry */ if (fdmi->retry_cnt++ < BFA_FCS_FDMI_CMD_MAX_RETRIES) { bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_rprt_retry); bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(port), &fdmi->timer, bfa_fcs_lport_fdmi_timeout, fdmi, BFA_FCS_RETRY_TIMEOUT); } else { /* * set state to offline */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); fdmi->retry_cnt = 0; } break; case FDMISM_EVENT_RSP_OK: fdmi->retry_cnt = 0; bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_online); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_fcxp_discard(fdmi->fcxp); bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_rprt_retry(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_sending_rprt); bfa_fcs_lport_fdmi_send_rprt(fdmi, NULL); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); bfa_timer_stop(&fdmi->timer); break; default: bfa_sm_fault(port->fcs, event); } } /* * Register Port Attributes */ static void bfa_fcs_lport_fdmi_sm_sending_rpa(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_RPA_SENT: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_rpa); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(port), &fdmi->fcxp_wqe); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_rpa(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_RSP_ERROR: /* * if max retries have not been reached, start timer for a * delayed retry */ if (fdmi->retry_cnt++ < BFA_FCS_FDMI_CMD_MAX_RETRIES) { bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_rpa_retry); bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(port), &fdmi->timer, bfa_fcs_lport_fdmi_timeout, fdmi, BFA_FCS_RETRY_TIMEOUT); } else { /* * set state to offline */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); fdmi->retry_cnt = 0; } break; case FDMISM_EVENT_RSP_OK: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_online); fdmi->retry_cnt = 0; break; case FDMISM_EVENT_PORT_OFFLINE: bfa_fcxp_discard(fdmi->fcxp); bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_rpa_retry(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_sending_rpa); bfa_fcs_lport_fdmi_send_rpa(fdmi, NULL); break; case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); bfa_timer_stop(&fdmi->timer); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_fdmi_sm_online(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); switch (event) { case FDMISM_EVENT_PORT_OFFLINE: bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); break; default: bfa_sm_fault(port->fcs, event); } } /* * FDMI is disabled state. */ static void bfa_fcs_lport_fdmi_sm_disabled(struct bfa_fcs_lport_fdmi_s *fdmi, enum port_fdmi_event event) { struct bfa_fcs_lport_s *port = fdmi->ms->port; bfa_trc(port->fcs, port->port_cfg.pwwn); bfa_trc(port->fcs, event); /* No op State. It can only be enabled at Driver Init. */ } /* * RHBA : Register HBA Attributes. */ static void bfa_fcs_lport_fdmi_send_rhba(void *fdmi_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_fdmi_s *fdmi = fdmi_cbarg; struct bfa_fcs_lport_s *port = fdmi->ms->port; struct fchs_s fchs; int len, attr_len; struct bfa_fcxp_s *fcxp; u8 *pyld; bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &fdmi->fcxp_wqe, bfa_fcs_lport_fdmi_send_rhba, fdmi); return; } fdmi->fcxp = fcxp; pyld = bfa_fcxp_get_reqbuf(fcxp); memset(pyld, 0, FC_MAX_PDUSZ); len = fc_fdmi_reqhdr_build(&fchs, pyld, bfa_fcs_lport_get_fcid(port), FDMI_RHBA); attr_len = bfa_fcs_lport_fdmi_build_rhba_pyld(fdmi, (u8 *) ((struct ct_hdr_s *) pyld + 1)); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, (len + attr_len), &fchs, bfa_fcs_lport_fdmi_rhba_response, (void *)fdmi, FC_MAX_PDUSZ, FC_FCCT_TOV); bfa_sm_send_event(fdmi, FDMISM_EVENT_RHBA_SENT); } static u16 bfa_fcs_lport_fdmi_build_rhba_pyld(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld) { struct bfa_fcs_lport_s *port = fdmi->ms->port; struct bfa_fcs_fdmi_hba_attr_s hba_attr; struct bfa_fcs_fdmi_hba_attr_s *fcs_hba_attr = &hba_attr; struct fdmi_rhba_s *rhba = (struct fdmi_rhba_s *) pyld; struct fdmi_attr_s *attr; u8 *curr_ptr; u16 len, count; u16 templen; /* * get hba attributes */ bfa_fcs_fdmi_get_hbaattr(fdmi, fcs_hba_attr); rhba->hba_id = bfa_fcs_lport_get_pwwn(port); rhba->port_list.num_ports = cpu_to_be32(1); rhba->port_list.port_entry = bfa_fcs_lport_get_pwwn(port); len = sizeof(rhba->hba_id) + sizeof(rhba->port_list); count = 0; len += sizeof(rhba->hba_attr_blk.attr_count); /* * fill out the invididual entries of the HBA attrib Block */ curr_ptr = (u8 *) &rhba->hba_attr_blk.hba_attr; /* * Node Name */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_NODENAME); templen = sizeof(wwn_t); memcpy(attr->value, &bfa_fcs_lport_get_nwwn(port), templen); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Manufacturer */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_MANUFACTURER); templen = (u16) strlen(fcs_hba_attr->manufacturer); memcpy(attr->value, fcs_hba_attr->manufacturer, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Serial Number */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_SERIALNUM); templen = (u16) strlen(fcs_hba_attr->serial_num); memcpy(attr->value, fcs_hba_attr->serial_num, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Model */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_MODEL); templen = (u16) strlen(fcs_hba_attr->model); memcpy(attr->value, fcs_hba_attr->model, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Model Desc */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_MODEL_DESC); templen = (u16) strlen(fcs_hba_attr->model_desc); memcpy(attr->value, fcs_hba_attr->model_desc, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * H/W Version */ if (fcs_hba_attr->hw_version[0] != '\0') { attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_HW_VERSION); templen = (u16) strlen(fcs_hba_attr->hw_version); memcpy(attr->value, fcs_hba_attr->hw_version, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); } /* * Driver Version */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_DRIVER_VERSION); templen = (u16) strlen(fcs_hba_attr->driver_version); memcpy(attr->value, fcs_hba_attr->driver_version, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Option Rom Version */ if (fcs_hba_attr->option_rom_ver[0] != '\0') { attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_ROM_VERSION); templen = (u16) strlen(fcs_hba_attr->option_rom_ver); memcpy(attr->value, fcs_hba_attr->option_rom_ver, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); } /* * f/w Version = driver version */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_FW_VERSION); templen = (u16) strlen(fcs_hba_attr->driver_version); memcpy(attr->value, fcs_hba_attr->driver_version, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * OS Name */ if (fcs_hba_attr->os_name[0] != '\0') { attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_OS_NAME); templen = (u16) strlen(fcs_hba_attr->os_name); memcpy(attr->value, fcs_hba_attr->os_name, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); } /* * MAX_CT_PAYLOAD */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_HBA_ATTRIB_MAX_CT); templen = sizeof(fcs_hba_attr->max_ct_pyld); memcpy(attr->value, &fcs_hba_attr->max_ct_pyld, templen); len += templen; count++; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Update size of payload */ len += ((sizeof(attr->type) + sizeof(attr->len)) * count); rhba->hba_attr_blk.attr_count = cpu_to_be32(count); return len; } static void bfa_fcs_lport_fdmi_rhba_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_fdmi_s *fdmi = (struct bfa_fcs_lport_fdmi_s *) cbarg; struct bfa_fcs_lport_s *port = fdmi->ms->port; struct ct_hdr_s *cthdr = NULL; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_OK); return; } bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_ERROR); } /* * RPRT : Register Port */ static void bfa_fcs_lport_fdmi_send_rprt(void *fdmi_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_fdmi_s *fdmi = fdmi_cbarg; struct bfa_fcs_lport_s *port = fdmi->ms->port; struct fchs_s fchs; u16 len, attr_len; struct bfa_fcxp_s *fcxp; u8 *pyld; bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &fdmi->fcxp_wqe, bfa_fcs_lport_fdmi_send_rprt, fdmi); return; } fdmi->fcxp = fcxp; pyld = bfa_fcxp_get_reqbuf(fcxp); memset(pyld, 0, FC_MAX_PDUSZ); len = fc_fdmi_reqhdr_build(&fchs, pyld, bfa_fcs_lport_get_fcid(port), FDMI_RPRT); attr_len = bfa_fcs_lport_fdmi_build_rprt_pyld(fdmi, (u8 *) ((struct ct_hdr_s *) pyld + 1)); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len + attr_len, &fchs, bfa_fcs_lport_fdmi_rprt_response, (void *)fdmi, FC_MAX_PDUSZ, FC_FCCT_TOV); bfa_sm_send_event(fdmi, FDMISM_EVENT_RPRT_SENT); } /* * This routine builds Port Attribute Block that used in RPA, RPRT commands. */ static u16 bfa_fcs_lport_fdmi_build_portattr_block(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld) { struct bfa_fcs_fdmi_port_attr_s fcs_port_attr; struct fdmi_port_attr_s *port_attrib = (struct fdmi_port_attr_s *) pyld; struct fdmi_attr_s *attr; u8 *curr_ptr; u16 len; u8 count = 0; u16 templen; /* * get port attributes */ bfa_fcs_fdmi_get_portattr(fdmi, &fcs_port_attr); len = sizeof(port_attrib->attr_count); /* * fill out the invididual entries */ curr_ptr = (u8 *) &port_attrib->port_attr; /* * FC4 Types */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_PORT_ATTRIB_FC4_TYPES); templen = sizeof(fcs_port_attr.supp_fc4_types); memcpy(attr->value, fcs_port_attr.supp_fc4_types, templen); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; ++count; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * Supported Speed */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_PORT_ATTRIB_SUPP_SPEED); templen = sizeof(fcs_port_attr.supp_speed); memcpy(attr->value, &fcs_port_attr.supp_speed, templen); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; ++count; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * current Port Speed */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_PORT_ATTRIB_PORT_SPEED); templen = sizeof(fcs_port_attr.curr_speed); memcpy(attr->value, &fcs_port_attr.curr_speed, templen); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; ++count; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * max frame size */ attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_PORT_ATTRIB_FRAME_SIZE); templen = sizeof(fcs_port_attr.max_frm_size); memcpy(attr->value, &fcs_port_attr.max_frm_size, templen); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; ++count; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); /* * OS Device Name */ if (fcs_port_attr.os_device_name[0] != '\0') { attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_PORT_ATTRIB_DEV_NAME); templen = (u16) strlen(fcs_port_attr.os_device_name); memcpy(attr->value, fcs_port_attr.os_device_name, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; ++count; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); } /* * Host Name */ if (fcs_port_attr.host_name[0] != '\0') { attr = (struct fdmi_attr_s *) curr_ptr; attr->type = cpu_to_be16(FDMI_PORT_ATTRIB_HOST_NAME); templen = (u16) strlen(fcs_port_attr.host_name); memcpy(attr->value, fcs_port_attr.host_name, templen); templen = fc_roundup(templen, sizeof(u32)); curr_ptr += sizeof(attr->type) + sizeof(templen) + templen; len += templen; ++count; attr->len = cpu_to_be16(templen + sizeof(attr->type) + sizeof(templen)); } /* * Update size of payload */ port_attrib->attr_count = cpu_to_be32(count); len += ((sizeof(attr->type) + sizeof(attr->len)) * count); return len; } static u16 bfa_fcs_lport_fdmi_build_rprt_pyld(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld) { struct bfa_fcs_lport_s *port = fdmi->ms->port; struct fdmi_rprt_s *rprt = (struct fdmi_rprt_s *) pyld; u16 len; rprt->hba_id = bfa_fcs_lport_get_pwwn(bfa_fcs_get_base_port(port->fcs)); rprt->port_name = bfa_fcs_lport_get_pwwn(port); len = bfa_fcs_lport_fdmi_build_portattr_block(fdmi, (u8 *) &rprt->port_attr_blk); len += sizeof(rprt->hba_id) + sizeof(rprt->port_name); return len; } static void bfa_fcs_lport_fdmi_rprt_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_fdmi_s *fdmi = (struct bfa_fcs_lport_fdmi_s *) cbarg; struct bfa_fcs_lport_s *port = fdmi->ms->port; struct ct_hdr_s *cthdr = NULL; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_OK); return; } bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_ERROR); } /* * RPA : Register Port Attributes. */ static void bfa_fcs_lport_fdmi_send_rpa(void *fdmi_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_fdmi_s *fdmi = fdmi_cbarg; struct bfa_fcs_lport_s *port = fdmi->ms->port; struct fchs_s fchs; u16 len, attr_len; struct bfa_fcxp_s *fcxp; u8 *pyld; bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &fdmi->fcxp_wqe, bfa_fcs_lport_fdmi_send_rpa, fdmi); return; } fdmi->fcxp = fcxp; pyld = bfa_fcxp_get_reqbuf(fcxp); memset(pyld, 0, FC_MAX_PDUSZ); len = fc_fdmi_reqhdr_build(&fchs, pyld, bfa_fcs_lport_get_fcid(port), FDMI_RPA); attr_len = bfa_fcs_lport_fdmi_build_rpa_pyld(fdmi, (u8 *) ((struct ct_hdr_s *) pyld + 1)); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len + attr_len, &fchs, bfa_fcs_lport_fdmi_rpa_response, (void *)fdmi, FC_MAX_PDUSZ, FC_FCCT_TOV); bfa_sm_send_event(fdmi, FDMISM_EVENT_RPA_SENT); } static u16 bfa_fcs_lport_fdmi_build_rpa_pyld(struct bfa_fcs_lport_fdmi_s *fdmi, u8 *pyld) { struct bfa_fcs_lport_s *port = fdmi->ms->port; struct fdmi_rpa_s *rpa = (struct fdmi_rpa_s *) pyld; u16 len; rpa->port_name = bfa_fcs_lport_get_pwwn(port); len = bfa_fcs_lport_fdmi_build_portattr_block(fdmi, (u8 *) &rpa->port_attr_blk); len += sizeof(rpa->port_name); return len; } static void bfa_fcs_lport_fdmi_rpa_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_fdmi_s *fdmi = (struct bfa_fcs_lport_fdmi_s *) cbarg; struct bfa_fcs_lport_s *port = fdmi->ms->port; struct ct_hdr_s *cthdr = NULL; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_OK); return; } bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(fdmi, FDMISM_EVENT_RSP_ERROR); } static void bfa_fcs_lport_fdmi_timeout(void *arg) { struct bfa_fcs_lport_fdmi_s *fdmi = (struct bfa_fcs_lport_fdmi_s *) arg; bfa_sm_send_event(fdmi, FDMISM_EVENT_TIMEOUT); } static void bfa_fcs_fdmi_get_hbaattr(struct bfa_fcs_lport_fdmi_s *fdmi, struct bfa_fcs_fdmi_hba_attr_s *hba_attr) { struct bfa_fcs_lport_s *port = fdmi->ms->port; struct bfa_fcs_driver_info_s *driver_info = &port->fcs->driver_info; memset(hba_attr, 0, sizeof(struct bfa_fcs_fdmi_hba_attr_s)); bfa_ioc_get_adapter_manufacturer(&port->fcs->bfa->ioc, hba_attr->manufacturer); bfa_ioc_get_adapter_serial_num(&port->fcs->bfa->ioc, hba_attr->serial_num); bfa_ioc_get_adapter_model(&port->fcs->bfa->ioc, hba_attr->model); bfa_ioc_get_adapter_model(&port->fcs->bfa->ioc, hba_attr->model_desc); bfa_ioc_get_pci_chip_rev(&port->fcs->bfa->ioc, hba_attr->hw_version); bfa_ioc_get_adapter_optrom_ver(&port->fcs->bfa->ioc, hba_attr->option_rom_ver); bfa_ioc_get_adapter_fw_ver(&port->fcs->bfa->ioc, hba_attr->fw_version); strncpy(hba_attr->driver_version, (char *)driver_info->version, sizeof(hba_attr->driver_version)); strncpy(hba_attr->os_name, driver_info->host_os_name, sizeof(hba_attr->os_name)); /* * If there is a patch level, append it * to the os name along with a separator */ if (driver_info->host_os_patch[0] != '\0') { strncat(hba_attr->os_name, BFA_FCS_PORT_SYMBNAME_SEPARATOR, sizeof(BFA_FCS_PORT_SYMBNAME_SEPARATOR)); strncat(hba_attr->os_name, driver_info->host_os_patch, sizeof(driver_info->host_os_patch)); } hba_attr->max_ct_pyld = cpu_to_be32(FC_MAX_PDUSZ); } static void bfa_fcs_fdmi_get_portattr(struct bfa_fcs_lport_fdmi_s *fdmi, struct bfa_fcs_fdmi_port_attr_s *port_attr) { struct bfa_fcs_lport_s *port = fdmi->ms->port; struct bfa_fcs_driver_info_s *driver_info = &port->fcs->driver_info; struct bfa_port_attr_s pport_attr; memset(port_attr, 0, sizeof(struct bfa_fcs_fdmi_port_attr_s)); /* * get pport attributes from hal */ bfa_fcport_get_attr(port->fcs->bfa, &pport_attr); /* * get FC4 type Bitmask */ fc_get_fc4type_bitmask(FC_TYPE_FCP, port_attr->supp_fc4_types); /* * Supported Speeds */ port_attr->supp_speed = cpu_to_be32(BFA_FCS_FDMI_SUPORTED_SPEEDS); /* * Current Speed */ port_attr->curr_speed = cpu_to_be32(pport_attr.speed); /* * Max PDU Size. */ port_attr->max_frm_size = cpu_to_be32(FC_MAX_PDUSZ); /* * OS device Name */ strncpy(port_attr->os_device_name, (char *)driver_info->os_device_name, sizeof(port_attr->os_device_name)); /* * Host name */ strncpy(port_attr->host_name, (char *)driver_info->host_machine_name, sizeof(port_attr->host_name)); } void bfa_fcs_lport_fdmi_init(struct bfa_fcs_lport_ms_s *ms) { struct bfa_fcs_lport_fdmi_s *fdmi = &ms->fdmi; fdmi->ms = ms; if (ms->port->fcs->fdmi_enabled) bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_offline); else bfa_sm_set_state(fdmi, bfa_fcs_lport_fdmi_sm_disabled); } void bfa_fcs_lport_fdmi_offline(struct bfa_fcs_lport_ms_s *ms) { struct bfa_fcs_lport_fdmi_s *fdmi = &ms->fdmi; fdmi->ms = ms; bfa_sm_send_event(fdmi, FDMISM_EVENT_PORT_OFFLINE); } void bfa_fcs_lport_fdmi_online(struct bfa_fcs_lport_ms_s *ms) { struct bfa_fcs_lport_fdmi_s *fdmi = &ms->fdmi; fdmi->ms = ms; bfa_sm_send_event(fdmi, FDMISM_EVENT_PORT_ONLINE); } #define BFA_FCS_MS_CMD_MAX_RETRIES 2 /* * forward declarations */ static void bfa_fcs_lport_ms_send_plogi(void *ms_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ms_timeout(void *arg); static void bfa_fcs_lport_ms_plogi_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ms_send_gmal(void *ms_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ms_gmal_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ms_send_gfn(void *ms_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ms_gfn_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); /* * fcs_ms_sm FCS MS state machine */ /* * MS State Machine events */ enum port_ms_event { MSSM_EVENT_PORT_ONLINE = 1, MSSM_EVENT_PORT_OFFLINE = 2, MSSM_EVENT_RSP_OK = 3, MSSM_EVENT_RSP_ERROR = 4, MSSM_EVENT_TIMEOUT = 5, MSSM_EVENT_FCXP_SENT = 6, MSSM_EVENT_PORT_FABRIC_RSCN = 7 }; static void bfa_fcs_lport_ms_sm_offline(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_plogi_sending(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_plogi(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_plogi_retry(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_gmal_sending(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_gmal(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_gmal_retry(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_gfn_sending(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_gfn(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_gfn_retry(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); static void bfa_fcs_lport_ms_sm_online(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event); /* * Start in offline state - awaiting NS to send start. */ static void bfa_fcs_lport_ms_sm_offline(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_PORT_ONLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_plogi_sending); bfa_fcs_lport_ms_send_plogi(ms, NULL); break; case MSSM_EVENT_PORT_OFFLINE: break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_plogi_sending(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_FCXP_SENT: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_plogi); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ms->port), &ms->fcxp_wqe); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_plogi(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_plogi_retry); ms->port->stats.ms_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ms->port), &ms->timer, bfa_fcs_lport_ms_timeout, ms, BFA_FCS_RETRY_TIMEOUT); break; case MSSM_EVENT_RSP_OK: /* * since plogi is done, now invoke MS related sub-modules */ bfa_fcs_lport_fdmi_online(ms); /* * if this is a Vport, go to online state. */ if (ms->port->vport) { bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_online); break; } /* * For a base port we need to get the * switch's IP address. */ bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gmal_sending); bfa_fcs_lport_ms_send_gmal(ms, NULL); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_fcxp_discard(ms->fcxp); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_plogi_retry(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_plogi_sending); bfa_fcs_lport_ms_send_plogi(ms, NULL); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_timer_stop(&ms->timer); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_online(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); break; case MSSM_EVENT_PORT_FABRIC_RSCN: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gfn_sending); ms->retry_cnt = 0; bfa_fcs_lport_ms_send_gfn(ms, NULL); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_gmal_sending(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_FCXP_SENT: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gmal); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ms->port), &ms->fcxp_wqe); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_gmal(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ if (ms->retry_cnt++ < BFA_FCS_MS_CMD_MAX_RETRIES) { bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gmal_retry); ms->port->stats.ms_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ms->port), &ms->timer, bfa_fcs_lport_ms_timeout, ms, BFA_FCS_RETRY_TIMEOUT); } else { bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gfn_sending); bfa_fcs_lport_ms_send_gfn(ms, NULL); ms->retry_cnt = 0; } break; case MSSM_EVENT_RSP_OK: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gfn_sending); bfa_fcs_lport_ms_send_gfn(ms, NULL); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_fcxp_discard(ms->fcxp); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_gmal_retry(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gmal_sending); bfa_fcs_lport_ms_send_gmal(ms, NULL); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_timer_stop(&ms->timer); break; default: bfa_sm_fault(ms->port->fcs, event); } } /* * ms_pvt MS local functions */ static void bfa_fcs_lport_ms_send_gmal(void *ms_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ms_s *ms = ms_cbarg; bfa_fcs_lport_t *port = ms->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->pid); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ms->fcxp_wqe, bfa_fcs_lport_ms_send_gmal, ms); return; } ms->fcxp = fcxp; len = fc_gmal_req_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_fcs_lport_get_fcid(port), port->fabric->lps->pr_nwwn); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ms_gmal_response, (void *)ms, FC_MAX_PDUSZ, FC_FCCT_TOV); bfa_sm_send_event(ms, MSSM_EVENT_FCXP_SENT); } static void bfa_fcs_lport_ms_gmal_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ms_s *ms = (struct bfa_fcs_lport_ms_s *) cbarg; bfa_fcs_lport_t *port = ms->port; struct ct_hdr_s *cthdr = NULL; struct fcgs_gmal_resp_s *gmal_resp; struct fcgs_gmal_entry_s *gmal_entry; u32 num_entries; u8 *rsp_str; bfa_trc(port->fcs, req_status); bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { gmal_resp = (struct fcgs_gmal_resp_s *)(cthdr + 1); num_entries = be32_to_cpu(gmal_resp->ms_len); if (num_entries == 0) { bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); return; } /* * The response could contain multiple Entries. * Entries for SNMP interface, etc. * We look for the entry with a telnet prefix. * First "http://" entry refers to IP addr */ gmal_entry = (struct fcgs_gmal_entry_s *)gmal_resp->ms_ma; while (num_entries > 0) { if (strncmp(gmal_entry->prefix, CT_GMAL_RESP_PREFIX_HTTP, sizeof(gmal_entry->prefix)) == 0) { /* * if the IP address is terminating with a '/', * remove it. * Byte 0 consists of the length of the string. */ rsp_str = &(gmal_entry->prefix[0]); if (rsp_str[gmal_entry->len-1] == '/') rsp_str[gmal_entry->len-1] = 0; /* copy IP Address to fabric */ strncpy(bfa_fcs_lport_get_fabric_ipaddr(port), gmal_entry->ip_addr, BFA_FCS_FABRIC_IPADDR_SZ); break; } else { --num_entries; ++gmal_entry; } } bfa_sm_send_event(ms, MSSM_EVENT_RSP_OK); return; } bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); } static void bfa_fcs_lport_ms_sm_gfn_sending(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_FCXP_SENT: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gfn); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ms->port), &ms->fcxp_wqe); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_gfn(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ if (ms->retry_cnt++ < BFA_FCS_MS_CMD_MAX_RETRIES) { bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gfn_retry); ms->port->stats.ms_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ms->port), &ms->timer, bfa_fcs_lport_ms_timeout, ms, BFA_FCS_RETRY_TIMEOUT); } else { bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_online); ms->retry_cnt = 0; } break; case MSSM_EVENT_RSP_OK: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_online); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_fcxp_discard(ms->fcxp); break; default: bfa_sm_fault(ms->port->fcs, event); } } static void bfa_fcs_lport_ms_sm_gfn_retry(struct bfa_fcs_lport_ms_s *ms, enum port_ms_event event) { bfa_trc(ms->port->fcs, ms->port->port_cfg.pwwn); bfa_trc(ms->port->fcs, event); switch (event) { case MSSM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_gfn_sending); bfa_fcs_lport_ms_send_gfn(ms, NULL); break; case MSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); bfa_timer_stop(&ms->timer); break; default: bfa_sm_fault(ms->port->fcs, event); } } /* * ms_pvt MS local functions */ static void bfa_fcs_lport_ms_send_gfn(void *ms_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ms_s *ms = ms_cbarg; bfa_fcs_lport_t *port = ms->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->pid); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ms->fcxp_wqe, bfa_fcs_lport_ms_send_gfn, ms); return; } ms->fcxp = fcxp; len = fc_gfn_req_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_fcs_lport_get_fcid(port), port->fabric->lps->pr_nwwn); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ms_gfn_response, (void *)ms, FC_MAX_PDUSZ, FC_FCCT_TOV); bfa_sm_send_event(ms, MSSM_EVENT_FCXP_SENT); } static void bfa_fcs_lport_ms_gfn_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ms_s *ms = (struct bfa_fcs_lport_ms_s *) cbarg; bfa_fcs_lport_t *port = ms->port; struct ct_hdr_s *cthdr = NULL; wwn_t *gfn_resp; bfa_trc(port->fcs, req_status); bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { gfn_resp = (wwn_t *)(cthdr + 1); /* check if it has actually changed */ if ((memcmp((void *)&bfa_fcs_lport_get_fabric_name(port), gfn_resp, sizeof(wwn_t)) != 0)) { bfa_fcs_fabric_set_fabric_name(port->fabric, *gfn_resp); } bfa_sm_send_event(ms, MSSM_EVENT_RSP_OK); return; } bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); } /* * ms_pvt MS local functions */ static void bfa_fcs_lport_ms_send_plogi(void *ms_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ms_s *ms = ms_cbarg; struct bfa_fcs_lport_s *port = ms->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->pid); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { port->stats.ms_plogi_alloc_wait++; bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ms->fcxp_wqe, bfa_fcs_lport_ms_send_plogi, ms); return; } ms->fcxp = fcxp; len = fc_plogi_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_hton3b(FC_MGMT_SERVER), bfa_fcs_lport_get_fcid(port), 0, port->port_cfg.pwwn, port->port_cfg.nwwn, bfa_fcport_get_maxfrsize(port->fcs->bfa), bfa_fcport_get_rx_bbcredit(port->fcs->bfa)); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ms_plogi_response, (void *)ms, FC_MAX_PDUSZ, FC_ELS_TOV); port->stats.ms_plogi_sent++; bfa_sm_send_event(ms, MSSM_EVENT_FCXP_SENT); } static void bfa_fcs_lport_ms_plogi_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ms_s *ms = (struct bfa_fcs_lport_ms_s *) cbarg; struct bfa_fcs_lport_s *port = ms->port; struct fc_els_cmd_s *els_cmd; struct fc_ls_rjt_s *ls_rjt; bfa_trc(port->fcs, req_status); bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { port->stats.ms_plogi_rsp_err++; bfa_trc(port->fcs, req_status); bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); return; } els_cmd = (struct fc_els_cmd_s *) BFA_FCXP_RSP_PLD(fcxp); switch (els_cmd->els_code) { case FC_ELS_ACC: if (rsp_len < sizeof(struct fc_logi_s)) { bfa_trc(port->fcs, rsp_len); port->stats.ms_plogi_acc_err++; bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); break; } port->stats.ms_plogi_accepts++; bfa_sm_send_event(ms, MSSM_EVENT_RSP_OK); break; case FC_ELS_LS_RJT: ls_rjt = (struct fc_ls_rjt_s *) BFA_FCXP_RSP_PLD(fcxp); bfa_trc(port->fcs, ls_rjt->reason_code); bfa_trc(port->fcs, ls_rjt->reason_code_expl); port->stats.ms_rejects++; bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); break; default: port->stats.ms_plogi_unknown_rsp++; bfa_trc(port->fcs, els_cmd->els_code); bfa_sm_send_event(ms, MSSM_EVENT_RSP_ERROR); } } static void bfa_fcs_lport_ms_timeout(void *arg) { struct bfa_fcs_lport_ms_s *ms = (struct bfa_fcs_lport_ms_s *) arg; ms->port->stats.ms_timeouts++; bfa_sm_send_event(ms, MSSM_EVENT_TIMEOUT); } void bfa_fcs_lport_ms_init(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ms_s *ms = BFA_FCS_GET_MS_FROM_PORT(port); ms->port = port; bfa_sm_set_state(ms, bfa_fcs_lport_ms_sm_offline); /* * Invoke init routines of sub modules. */ bfa_fcs_lport_fdmi_init(ms); } void bfa_fcs_lport_ms_offline(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ms_s *ms = BFA_FCS_GET_MS_FROM_PORT(port); ms->port = port; bfa_sm_send_event(ms, MSSM_EVENT_PORT_OFFLINE); bfa_fcs_lport_fdmi_offline(ms); } void bfa_fcs_lport_ms_online(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ms_s *ms = BFA_FCS_GET_MS_FROM_PORT(port); ms->port = port; bfa_sm_send_event(ms, MSSM_EVENT_PORT_ONLINE); } void bfa_fcs_lport_ms_fabric_rscn(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ms_s *ms = BFA_FCS_GET_MS_FROM_PORT(port); /* todo. Handle this only when in Online state */ if (bfa_sm_cmp_state(ms, bfa_fcs_lport_ms_sm_online)) bfa_sm_send_event(ms, MSSM_EVENT_PORT_FABRIC_RSCN); } /* * @page ns_sm_info VPORT NS State Machine * * @section ns_sm_interactions VPORT NS State Machine Interactions * * @section ns_sm VPORT NS State Machine * img ns_sm.jpg */ /* * forward declarations */ static void bfa_fcs_lport_ns_send_plogi(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ns_send_rspn_id(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ns_send_rft_id(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ns_send_rff_id(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ns_send_gid_ft(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_ns_timeout(void *arg); static void bfa_fcs_lport_ns_plogi_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ns_rspn_id_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ns_rft_id_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ns_rff_id_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ns_gid_ft_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_ns_process_gidft_pids( struct bfa_fcs_lport_s *port, u32 *pid_buf, u32 n_pids); static void bfa_fcs_lport_ns_boot_target_disc(bfa_fcs_lport_t *port); /* * fcs_ns_sm FCS nameserver interface state machine */ /* * VPort NS State Machine events */ enum vport_ns_event { NSSM_EVENT_PORT_ONLINE = 1, NSSM_EVENT_PORT_OFFLINE = 2, NSSM_EVENT_PLOGI_SENT = 3, NSSM_EVENT_RSP_OK = 4, NSSM_EVENT_RSP_ERROR = 5, NSSM_EVENT_TIMEOUT = 6, NSSM_EVENT_NS_QUERY = 7, NSSM_EVENT_RSPNID_SENT = 8, NSSM_EVENT_RFTID_SENT = 9, NSSM_EVENT_RFFID_SENT = 10, NSSM_EVENT_GIDFT_SENT = 11, }; static void bfa_fcs_lport_ns_sm_offline(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_plogi_sending(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_plogi(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_plogi_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_sending_rspn_id( struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_rspn_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_rspn_id_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_sending_rft_id( struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_rft_id_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_rft_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_sending_rff_id( struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_rff_id_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_rff_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_sending_gid_ft( struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_gid_ft(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_gid_ft_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); static void bfa_fcs_lport_ns_sm_online(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event); /* * Start in offline state - awaiting linkup */ static void bfa_fcs_lport_ns_sm_offline(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_PORT_ONLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_plogi_sending); bfa_fcs_lport_ns_send_plogi(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_plogi_sending(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_PLOGI_SENT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_plogi); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->fcxp_wqe); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_plogi(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_plogi_retry); ns->port->stats.ns_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->timer, bfa_fcs_lport_ns_timeout, ns, BFA_FCS_RETRY_TIMEOUT); break; case NSSM_EVENT_RSP_OK: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_rspn_id); bfa_fcs_lport_ns_send_rspn_id(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_discard(ns->fcxp); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_plogi_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_plogi_sending); bfa_fcs_lport_ns_send_plogi(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_timer_stop(&ns->timer); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_sending_rspn_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RSPNID_SENT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_rspn_id); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->fcxp_wqe); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_rspn_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_rspn_id_retry); ns->port->stats.ns_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->timer, bfa_fcs_lport_ns_timeout, ns, BFA_FCS_RETRY_TIMEOUT); break; case NSSM_EVENT_RSP_OK: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_rft_id); bfa_fcs_lport_ns_send_rft_id(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_fcxp_discard(ns->fcxp); bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_rspn_id_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_TIMEOUT: /* * Retry Timer Expired. Re-send */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_rspn_id); bfa_fcs_lport_ns_send_rspn_id(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_timer_stop(&ns->timer); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_sending_rft_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RFTID_SENT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_rft_id); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->fcxp_wqe); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_rft_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RSP_OK: /* Now move to register FC4 Features */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_rff_id); bfa_fcs_lport_ns_send_rff_id(ns, NULL); break; case NSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_rft_id_retry); ns->port->stats.ns_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->timer, bfa_fcs_lport_ns_timeout, ns, BFA_FCS_RETRY_TIMEOUT); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_discard(ns->fcxp); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_rft_id_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_TIMEOUT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_rft_id); bfa_fcs_lport_ns_send_rft_id(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_timer_stop(&ns->timer); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_sending_rff_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RFFID_SENT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_rff_id); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->fcxp_wqe); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_rff_id(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RSP_OK: /* * If min cfg mode is enabled, we donot initiate rport * discovery with the fabric. Instead, we will retrieve the * boot targets from HAL/FW. */ if (__fcs_min_cfg(ns->port->fcs)) { bfa_fcs_lport_ns_boot_target_disc(ns->port); bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_online); return; } /* * If the port role is Initiator Mode issue NS query. * If it is Target Mode, skip this and go to online. */ if (BFA_FCS_VPORT_IS_INITIATOR_MODE(ns->port)) { bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_gid_ft); bfa_fcs_lport_ns_send_gid_ft(ns, NULL); } /* * kick off mgmt srvr state machine */ bfa_fcs_lport_ms_online(ns->port); break; case NSSM_EVENT_RSP_ERROR: /* * Start timer for a delayed retry */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_rff_id_retry); ns->port->stats.ns_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->timer, bfa_fcs_lport_ns_timeout, ns, BFA_FCS_RETRY_TIMEOUT); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_discard(ns->fcxp); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_rff_id_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_TIMEOUT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_rff_id); bfa_fcs_lport_ns_send_rff_id(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_timer_stop(&ns->timer); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_sending_gid_ft(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_GIDFT_SENT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_gid_ft); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_walloc_cancel(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->fcxp_wqe); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_gid_ft(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_RSP_OK: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_online); break; case NSSM_EVENT_RSP_ERROR: /* * TBD: for certain reject codes, we don't need to retry */ /* * Start timer for a delayed retry */ bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_gid_ft_retry); ns->port->stats.ns_retries++; bfa_timer_start(BFA_FCS_GET_HAL_FROM_PORT(ns->port), &ns->timer, bfa_fcs_lport_ns_timeout, ns, BFA_FCS_RETRY_TIMEOUT); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_fcxp_discard(ns->fcxp); break; case NSSM_EVENT_NS_QUERY: break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_gid_ft_retry(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_TIMEOUT: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_gid_ft); bfa_fcs_lport_ns_send_gid_ft(ns, NULL); break; case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); bfa_timer_stop(&ns->timer); break; default: bfa_sm_fault(ns->port->fcs, event); } } static void bfa_fcs_lport_ns_sm_online(struct bfa_fcs_lport_ns_s *ns, enum vport_ns_event event) { bfa_trc(ns->port->fcs, ns->port->port_cfg.pwwn); bfa_trc(ns->port->fcs, event); switch (event) { case NSSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); break; case NSSM_EVENT_NS_QUERY: /* * If the port role is Initiator Mode issue NS query. * If it is Target Mode, skip this and go to online. */ if (BFA_FCS_VPORT_IS_INITIATOR_MODE(ns->port)) { bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_sending_gid_ft); bfa_fcs_lport_ns_send_gid_ft(ns, NULL); }; break; default: bfa_sm_fault(ns->port->fcs, event); } } /* * ns_pvt Nameserver local functions */ static void bfa_fcs_lport_ns_send_plogi(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ns_s *ns = ns_cbarg; struct bfa_fcs_lport_s *port = ns->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->pid); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { port->stats.ns_plogi_alloc_wait++; bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ns->fcxp_wqe, bfa_fcs_lport_ns_send_plogi, ns); return; } ns->fcxp = fcxp; len = fc_plogi_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_hton3b(FC_NAME_SERVER), bfa_fcs_lport_get_fcid(port), 0, port->port_cfg.pwwn, port->port_cfg.nwwn, bfa_fcport_get_maxfrsize(port->fcs->bfa), bfa_fcport_get_rx_bbcredit(port->fcs->bfa)); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ns_plogi_response, (void *)ns, FC_MAX_PDUSZ, FC_ELS_TOV); port->stats.ns_plogi_sent++; bfa_sm_send_event(ns, NSSM_EVENT_PLOGI_SENT); } static void bfa_fcs_lport_ns_plogi_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ns_s *ns = (struct bfa_fcs_lport_ns_s *) cbarg; struct bfa_fcs_lport_s *port = ns->port; /* struct fc_logi_s *plogi_resp; */ struct fc_els_cmd_s *els_cmd; struct fc_ls_rjt_s *ls_rjt; bfa_trc(port->fcs, req_status); bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); port->stats.ns_plogi_rsp_err++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); return; } els_cmd = (struct fc_els_cmd_s *) BFA_FCXP_RSP_PLD(fcxp); switch (els_cmd->els_code) { case FC_ELS_ACC: if (rsp_len < sizeof(struct fc_logi_s)) { bfa_trc(port->fcs, rsp_len); port->stats.ns_plogi_acc_err++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); break; } port->stats.ns_plogi_accepts++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); break; case FC_ELS_LS_RJT: ls_rjt = (struct fc_ls_rjt_s *) BFA_FCXP_RSP_PLD(fcxp); bfa_trc(port->fcs, ls_rjt->reason_code); bfa_trc(port->fcs, ls_rjt->reason_code_expl); port->stats.ns_rejects++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); break; default: port->stats.ns_plogi_unknown_rsp++; bfa_trc(port->fcs, els_cmd->els_code); bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); } } /* * Register the symbolic port name. */ static void bfa_fcs_lport_ns_send_rspn_id(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ns_s *ns = ns_cbarg; struct bfa_fcs_lport_s *port = ns->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; u8 symbl[256]; u8 *psymbl = &symbl[0]; memset(symbl, 0, sizeof(symbl)); bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { port->stats.ns_rspnid_alloc_wait++; bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ns->fcxp_wqe, bfa_fcs_lport_ns_send_rspn_id, ns); return; } ns->fcxp = fcxp; /* * for V-Port, form a Port Symbolic Name */ if (port->vport) { /* * For Vports, we append the vport's port symbolic name * to that of the base port. */ strncpy((char *)psymbl, (char *) & (bfa_fcs_lport_get_psym_name (bfa_fcs_get_base_port(port->fcs))), strlen((char *) & bfa_fcs_lport_get_psym_name(bfa_fcs_get_base_port (port->fcs)))); /* Ensure we have a null terminating string. */ ((char *)psymbl)[strlen((char *) & bfa_fcs_lport_get_psym_name(bfa_fcs_get_base_port (port->fcs)))] = 0; strncat((char *)psymbl, (char *) &(bfa_fcs_lport_get_psym_name(port)), strlen((char *) &bfa_fcs_lport_get_psym_name(port))); } else { psymbl = (u8 *) &(bfa_fcs_lport_get_psym_name(port)); } len = fc_rspnid_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_fcs_lport_get_fcid(port), 0, psymbl); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ns_rspn_id_response, (void *)ns, FC_MAX_PDUSZ, FC_FCCT_TOV); port->stats.ns_rspnid_sent++; bfa_sm_send_event(ns, NSSM_EVENT_RSPNID_SENT); } static void bfa_fcs_lport_ns_rspn_id_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ns_s *ns = (struct bfa_fcs_lport_ns_s *) cbarg; struct bfa_fcs_lport_s *port = ns->port; struct ct_hdr_s *cthdr = NULL; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); port->stats.ns_rspnid_rsp_err++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { port->stats.ns_rspnid_accepts++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); return; } port->stats.ns_rspnid_rejects++; bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); } /* * Register FC4-Types */ static void bfa_fcs_lport_ns_send_rft_id(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ns_s *ns = ns_cbarg; struct bfa_fcs_lport_s *port = ns->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { port->stats.ns_rftid_alloc_wait++; bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ns->fcxp_wqe, bfa_fcs_lport_ns_send_rft_id, ns); return; } ns->fcxp = fcxp; len = fc_rftid_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_fcs_lport_get_fcid(port), 0, port->port_cfg.roles); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ns_rft_id_response, (void *)ns, FC_MAX_PDUSZ, FC_FCCT_TOV); port->stats.ns_rftid_sent++; bfa_sm_send_event(ns, NSSM_EVENT_RFTID_SENT); } static void bfa_fcs_lport_ns_rft_id_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ns_s *ns = (struct bfa_fcs_lport_ns_s *) cbarg; struct bfa_fcs_lport_s *port = ns->port; struct ct_hdr_s *cthdr = NULL; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); port->stats.ns_rftid_rsp_err++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { port->stats.ns_rftid_accepts++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); return; } port->stats.ns_rftid_rejects++; bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); } /* * Register FC4-Features : Should be done after RFT_ID */ static void bfa_fcs_lport_ns_send_rff_id(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ns_s *ns = ns_cbarg; struct bfa_fcs_lport_s *port = ns->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; u8 fc4_ftrs = 0; bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { port->stats.ns_rffid_alloc_wait++; bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ns->fcxp_wqe, bfa_fcs_lport_ns_send_rff_id, ns); return; } ns->fcxp = fcxp; if (BFA_FCS_VPORT_IS_INITIATOR_MODE(ns->port)) fc4_ftrs = FC_GS_FCP_FC4_FEATURE_INITIATOR; len = fc_rffid_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), bfa_fcs_lport_get_fcid(port), 0, FC_TYPE_FCP, fc4_ftrs); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ns_rff_id_response, (void *)ns, FC_MAX_PDUSZ, FC_FCCT_TOV); port->stats.ns_rffid_sent++; bfa_sm_send_event(ns, NSSM_EVENT_RFFID_SENT); } static void bfa_fcs_lport_ns_rff_id_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ns_s *ns = (struct bfa_fcs_lport_ns_s *) cbarg; struct bfa_fcs_lport_s *port = ns->port; struct ct_hdr_s *cthdr = NULL; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); port->stats.ns_rffid_rsp_err++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); if (cthdr->cmd_rsp_code == CT_RSP_ACCEPT) { port->stats.ns_rffid_accepts++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); return; } port->stats.ns_rffid_rejects++; bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); if (cthdr->reason_code == CT_RSN_NOT_SUPP) { /* if this command is not supported, we don't retry */ bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); } else bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); } /* * Query Fabric for FC4-Types Devices. * * TBD : Need to use a local (FCS private) response buffer, since the response * can be larger than 2K. */ static void bfa_fcs_lport_ns_send_gid_ft(void *ns_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_ns_s *ns = ns_cbarg; struct bfa_fcs_lport_s *port = ns->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->pid); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { port->stats.ns_gidft_alloc_wait++; bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &ns->fcxp_wqe, bfa_fcs_lport_ns_send_gid_ft, ns); return; } ns->fcxp = fcxp; /* * This query is only initiated for FCP initiator mode. */ len = fc_gid_ft_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), ns->port->pid, FC_TYPE_FCP); bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_ns_gid_ft_response, (void *)ns, bfa_fcxp_get_maxrsp(port->fcs->bfa), FC_FCCT_TOV); port->stats.ns_gidft_sent++; bfa_sm_send_event(ns, NSSM_EVENT_GIDFT_SENT); } static void bfa_fcs_lport_ns_gid_ft_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_ns_s *ns = (struct bfa_fcs_lport_ns_s *) cbarg; struct bfa_fcs_lport_s *port = ns->port; struct ct_hdr_s *cthdr = NULL; u32 n_pids; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); port->stats.ns_gidft_rsp_err++; bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); return; } if (resid_len != 0) { /* * TBD : we will need to allocate a larger buffer & retry the * command */ bfa_trc(port->fcs, rsp_len); bfa_trc(port->fcs, resid_len); return; } cthdr = (struct ct_hdr_s *) BFA_FCXP_RSP_PLD(fcxp); cthdr->cmd_rsp_code = be16_to_cpu(cthdr->cmd_rsp_code); switch (cthdr->cmd_rsp_code) { case CT_RSP_ACCEPT: port->stats.ns_gidft_accepts++; n_pids = (fc_get_ctresp_pyld_len(rsp_len) / sizeof(u32)); bfa_trc(port->fcs, n_pids); bfa_fcs_lport_ns_process_gidft_pids(port, (u32 *) (cthdr + 1), n_pids); bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); break; case CT_RSP_REJECT: /* * Check the reason code & explanation. * There may not have been any FC4 devices in the fabric */ port->stats.ns_gidft_rejects++; bfa_trc(port->fcs, cthdr->reason_code); bfa_trc(port->fcs, cthdr->exp_code); if ((cthdr->reason_code == CT_RSN_UNABLE_TO_PERF) && (cthdr->exp_code == CT_NS_EXP_FT_NOT_REG)) { bfa_sm_send_event(ns, NSSM_EVENT_RSP_OK); } else { /* * for all other errors, retry */ bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); } break; default: port->stats.ns_gidft_unknown_rsp++; bfa_trc(port->fcs, cthdr->cmd_rsp_code); bfa_sm_send_event(ns, NSSM_EVENT_RSP_ERROR); } } /* * This routine will be called by bfa_timer on timer timeouts. * * param[in] port - pointer to bfa_fcs_lport_t. * * return * void * * Special Considerations: * * note */ static void bfa_fcs_lport_ns_timeout(void *arg) { struct bfa_fcs_lport_ns_s *ns = (struct bfa_fcs_lport_ns_s *) arg; ns->port->stats.ns_timeouts++; bfa_sm_send_event(ns, NSSM_EVENT_TIMEOUT); } /* * Process the PID list in GID_FT response */ static void bfa_fcs_lport_ns_process_gidft_pids(struct bfa_fcs_lport_s *port, u32 *pid_buf, u32 n_pids) { struct fcgs_gidft_resp_s *gidft_entry; struct bfa_fcs_rport_s *rport; u32 ii; for (ii = 0; ii < n_pids; ii++) { gidft_entry = (struct fcgs_gidft_resp_s *) &pid_buf[ii]; if (gidft_entry->pid == port->pid) continue; /* * Check if this rport already exists */ rport = bfa_fcs_lport_get_rport_by_pid(port, gidft_entry->pid); if (rport == NULL) { /* * this is a new device. create rport */ rport = bfa_fcs_rport_create(port, gidft_entry->pid); } else { /* * this rport already exists */ bfa_fcs_rport_scn(rport); } bfa_trc(port->fcs, gidft_entry->pid); /* * if the last entry bit is set, bail out. */ if (gidft_entry->last) return; } } /* * fcs_ns_public FCS nameserver public interfaces */ /* * Functions called by port/fab. * These will send relevant Events to the ns state machine. */ void bfa_fcs_lport_ns_init(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ns_s *ns = BFA_FCS_GET_NS_FROM_PORT(port); ns->port = port; bfa_sm_set_state(ns, bfa_fcs_lport_ns_sm_offline); } void bfa_fcs_lport_ns_offline(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ns_s *ns = BFA_FCS_GET_NS_FROM_PORT(port); ns->port = port; bfa_sm_send_event(ns, NSSM_EVENT_PORT_OFFLINE); } void bfa_fcs_lport_ns_online(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ns_s *ns = BFA_FCS_GET_NS_FROM_PORT(port); ns->port = port; bfa_sm_send_event(ns, NSSM_EVENT_PORT_ONLINE); } void bfa_fcs_lport_ns_query(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_ns_s *ns = BFA_FCS_GET_NS_FROM_PORT(port); bfa_trc(port->fcs, port->pid); bfa_sm_send_event(ns, NSSM_EVENT_NS_QUERY); } static void bfa_fcs_lport_ns_boot_target_disc(bfa_fcs_lport_t *port) { struct bfa_fcs_rport_s *rport; u8 nwwns; wwn_t wwns[BFA_PREBOOT_BOOTLUN_MAX]; int ii; bfa_iocfc_get_bootwwns(port->fcs->bfa, &nwwns, wwns); for (ii = 0 ; ii < nwwns; ++ii) { rport = bfa_fcs_rport_create_by_wwn(port, wwns[ii]); WARN_ON(!rport); } } /* * FCS SCN */ #define FC_QOS_RSCN_EVENT 0x0c #define FC_FABRIC_NAME_RSCN_EVENT 0x0d /* * forward declarations */ static void bfa_fcs_lport_scn_send_scr(void *scn_cbarg, struct bfa_fcxp_s *fcxp_alloced); static void bfa_fcs_lport_scn_scr_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs); static void bfa_fcs_lport_scn_send_ls_acc(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs); static void bfa_fcs_lport_scn_timeout(void *arg); /* * fcs_scm_sm FCS SCN state machine */ /* * VPort SCN State Machine events */ enum port_scn_event { SCNSM_EVENT_PORT_ONLINE = 1, SCNSM_EVENT_PORT_OFFLINE = 2, SCNSM_EVENT_RSP_OK = 3, SCNSM_EVENT_RSP_ERROR = 4, SCNSM_EVENT_TIMEOUT = 5, SCNSM_EVENT_SCR_SENT = 6, }; static void bfa_fcs_lport_scn_sm_offline(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event); static void bfa_fcs_lport_scn_sm_sending_scr( struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event); static void bfa_fcs_lport_scn_sm_scr(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event); static void bfa_fcs_lport_scn_sm_scr_retry(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event); static void bfa_fcs_lport_scn_sm_online(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event); /* * Starting state - awaiting link up. */ static void bfa_fcs_lport_scn_sm_offline(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event) { switch (event) { case SCNSM_EVENT_PORT_ONLINE: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_sending_scr); bfa_fcs_lport_scn_send_scr(scn, NULL); break; case SCNSM_EVENT_PORT_OFFLINE: break; default: bfa_sm_fault(scn->port->fcs, event); } } static void bfa_fcs_lport_scn_sm_sending_scr(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event) { switch (event) { case SCNSM_EVENT_SCR_SENT: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_scr); break; case SCNSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_offline); bfa_fcxp_walloc_cancel(scn->port->fcs->bfa, &scn->fcxp_wqe); break; default: bfa_sm_fault(scn->port->fcs, event); } } static void bfa_fcs_lport_scn_sm_scr(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event) { struct bfa_fcs_lport_s *port = scn->port; switch (event) { case SCNSM_EVENT_RSP_OK: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_online); break; case SCNSM_EVENT_RSP_ERROR: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_scr_retry); bfa_timer_start(port->fcs->bfa, &scn->timer, bfa_fcs_lport_scn_timeout, scn, BFA_FCS_RETRY_TIMEOUT); break; case SCNSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_offline); bfa_fcxp_discard(scn->fcxp); break; default: bfa_sm_fault(port->fcs, event); } } static void bfa_fcs_lport_scn_sm_scr_retry(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event) { switch (event) { case SCNSM_EVENT_TIMEOUT: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_sending_scr); bfa_fcs_lport_scn_send_scr(scn, NULL); break; case SCNSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_offline); bfa_timer_stop(&scn->timer); break; default: bfa_sm_fault(scn->port->fcs, event); } } static void bfa_fcs_lport_scn_sm_online(struct bfa_fcs_lport_scn_s *scn, enum port_scn_event event) { switch (event) { case SCNSM_EVENT_PORT_OFFLINE: bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_offline); break; default: bfa_sm_fault(scn->port->fcs, event); } } /* * fcs_scn_private FCS SCN private functions */ /* * This routine will be called to send a SCR command. */ static void bfa_fcs_lport_scn_send_scr(void *scn_cbarg, struct bfa_fcxp_s *fcxp_alloced) { struct bfa_fcs_lport_scn_s *scn = scn_cbarg; struct bfa_fcs_lport_s *port = scn->port; struct fchs_s fchs; int len; struct bfa_fcxp_s *fcxp; bfa_trc(port->fcs, port->pid); bfa_trc(port->fcs, port->port_cfg.pwwn); fcxp = fcxp_alloced ? fcxp_alloced : bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) { bfa_fcs_fcxp_alloc_wait(port->fcs->bfa, &scn->fcxp_wqe, bfa_fcs_lport_scn_send_scr, scn); return; } scn->fcxp = fcxp; /* Handle VU registrations for Base port only */ if ((!port->vport) && bfa_ioc_get_fcmode(&port->fcs->bfa->ioc)) { len = fc_scr_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), port->fabric->lps->brcd_switch, port->pid, 0); } else { len = fc_scr_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), BFA_FALSE, port->pid, 0); } bfa_fcxp_send(fcxp, NULL, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, bfa_fcs_lport_scn_scr_response, (void *)scn, FC_MAX_PDUSZ, FC_ELS_TOV); bfa_sm_send_event(scn, SCNSM_EVENT_SCR_SENT); } static void bfa_fcs_lport_scn_scr_response(void *fcsarg, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { struct bfa_fcs_lport_scn_s *scn = (struct bfa_fcs_lport_scn_s *) cbarg; struct bfa_fcs_lport_s *port = scn->port; struct fc_els_cmd_s *els_cmd; struct fc_ls_rjt_s *ls_rjt; bfa_trc(port->fcs, port->port_cfg.pwwn); /* * Sanity Checks */ if (req_status != BFA_STATUS_OK) { bfa_trc(port->fcs, req_status); bfa_sm_send_event(scn, SCNSM_EVENT_RSP_ERROR); return; } els_cmd = (struct fc_els_cmd_s *) BFA_FCXP_RSP_PLD(fcxp); switch (els_cmd->els_code) { case FC_ELS_ACC: bfa_sm_send_event(scn, SCNSM_EVENT_RSP_OK); break; case FC_ELS_LS_RJT: ls_rjt = (struct fc_ls_rjt_s *) BFA_FCXP_RSP_PLD(fcxp); bfa_trc(port->fcs, ls_rjt->reason_code); bfa_trc(port->fcs, ls_rjt->reason_code_expl); bfa_sm_send_event(scn, SCNSM_EVENT_RSP_ERROR); break; default: bfa_sm_send_event(scn, SCNSM_EVENT_RSP_ERROR); } } /* * Send a LS Accept */ static void bfa_fcs_lport_scn_send_ls_acc(struct bfa_fcs_lport_s *port, struct fchs_s *rx_fchs) { struct fchs_s fchs; struct bfa_fcxp_s *fcxp; struct bfa_rport_s *bfa_rport = NULL; int len; bfa_trc(port->fcs, rx_fchs->s_id); fcxp = bfa_fcs_fcxp_alloc(port->fcs); if (!fcxp) return; len = fc_ls_acc_build(&fchs, bfa_fcxp_get_reqbuf(fcxp), rx_fchs->s_id, bfa_fcs_lport_get_fcid(port), rx_fchs->ox_id); bfa_fcxp_send(fcxp, bfa_rport, port->fabric->vf_id, port->lp_tag, BFA_FALSE, FC_CLASS_3, len, &fchs, NULL, NULL, FC_MAX_PDUSZ, 0); } /* * This routine will be called by bfa_timer on timer timeouts. * * param[in] vport - pointer to bfa_fcs_lport_t. * param[out] vport_status - pointer to return vport status in * * return * void * * Special Considerations: * * note */ static void bfa_fcs_lport_scn_timeout(void *arg) { struct bfa_fcs_lport_scn_s *scn = (struct bfa_fcs_lport_scn_s *) arg; bfa_sm_send_event(scn, SCNSM_EVENT_TIMEOUT); } /* * fcs_scn_public FCS state change notification public interfaces */ /* * Functions called by port/fab */ void bfa_fcs_lport_scn_init(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_scn_s *scn = BFA_FCS_GET_SCN_FROM_PORT(port); scn->port = port; bfa_sm_set_state(scn, bfa_fcs_lport_scn_sm_offline); } void bfa_fcs_lport_scn_offline(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_scn_s *scn = BFA_FCS_GET_SCN_FROM_PORT(port); scn->port = port; bfa_sm_send_event(scn, SCNSM_EVENT_PORT_OFFLINE); } void bfa_fcs_lport_scn_online(struct bfa_fcs_lport_s *port) { struct bfa_fcs_lport_scn_s *scn = BFA_FCS_GET_SCN_FROM_PORT(port); scn->port = port; bfa_sm_send_event(scn, SCNSM_EVENT_PORT_ONLINE); } static void bfa_fcs_lport_scn_portid_rscn(struct bfa_fcs_lport_s *port, u32 rpid) { struct bfa_fcs_rport_s *rport; bfa_trc(port->fcs, rpid); /* * If this is an unknown device, then it just came online. * Otherwise let rport handle the RSCN event. */ rport = bfa_fcs_lport_get_rport_by_pid(port, rpid); if (rport == NULL) { /* * If min cfg mode is enabled, we donot need to * discover any new rports. */ if (!__fcs_min_cfg(port->fcs)) rport = bfa_fcs_rport_create(port, rpid); } else bfa_fcs_rport_scn(rport); } /* * rscn format based PID comparison */ #define __fc_pid_match(__c0, __c1, __fmt) \ (((__fmt) == FC_RSCN_FORMAT_FABRIC) || \ (((__fmt) == FC_RSCN_FORMAT_DOMAIN) && \ ((__c0)[0] == (__c1)[0])) || \ (((__fmt) == FC_RSCN_FORMAT_AREA) && \ ((__c0)[0] == (__c1)[0]) && \ ((__c0)[1] == (__c1)[1]))) static void bfa_fcs_lport_scn_multiport_rscn(struct bfa_fcs_lport_s *port, enum fc_rscn_format format, u32 rscn_pid) { struct bfa_fcs_rport_s *rport; struct list_head *qe, *qe_next; u8 *c0, *c1; bfa_trc(port->fcs, format); bfa_trc(port->fcs, rscn_pid); c0 = (u8 *) &rscn_pid; list_for_each_safe(qe, qe_next, &port->rport_q) { rport = (struct bfa_fcs_rport_s *) qe; c1 = (u8 *) &rport->pid; if (__fc_pid_match(c0, c1, format)) bfa_fcs_rport_scn(rport); } } void bfa_fcs_lport_scn_process_rscn(struct bfa_fcs_lport_s *port, struct fchs_s *fchs, u32 len) { struct fc_rscn_pl_s *rscn = (struct fc_rscn_pl_s *) (fchs + 1); int num_entries; u32 rscn_pid; bfa_boolean_t nsquery = BFA_FALSE, found; int i = 0, j; num_entries = (be16_to_cpu(rscn->payldlen) - sizeof(u32)) / sizeof(rscn->event[0]); bfa_trc(port->fcs, num_entries); port->stats.num_rscn++; bfa_fcs_lport_scn_send_ls_acc(port, fchs); for (i = 0; i < num_entries; i++) { rscn_pid = rscn->event[i].portid; bfa_trc(port->fcs, rscn->event[i].format); bfa_trc(port->fcs, rscn_pid); /* check for duplicate entries in the list */ found = BFA_FALSE; for (j = 0; j < i; j++) { if (rscn->event[j].portid == rscn_pid) { found = BFA_TRUE; break; } } /* if found in down the list, pid has been already processed */ if (found) { bfa_trc(port->fcs, rscn_pid); continue; } switch (rscn->event[i].format) { case FC_RSCN_FORMAT_PORTID: if (rscn->event[i].qualifier == FC_QOS_RSCN_EVENT) { /* * Ignore this event. * f/w would have processed it */ bfa_trc(port->fcs, rscn_pid); } else { port->stats.num_portid_rscn++; bfa_fcs_lport_scn_portid_rscn(port, rscn_pid); } break; case FC_RSCN_FORMAT_FABRIC: if (rscn->event[i].qualifier == FC_FABRIC_NAME_RSCN_EVENT) { bfa_fcs_lport_ms_fabric_rscn(port); break; } /* !!!!!!!!! Fall Through !!!!!!!!!!!!! */ case FC_RSCN_FORMAT_AREA: case FC_RSCN_FORMAT_DOMAIN: nsquery = BFA_TRUE; bfa_fcs_lport_scn_multiport_rscn(port, rscn->event[i].format, rscn_pid); break; default: WARN_ON(1); nsquery = BFA_TRUE; } } /* * If any of area, domain or fabric RSCN is received, do a fresh * discovery to find new devices. */ if (nsquery) bfa_fcs_lport_ns_query(port); } /* * BFA FCS port */ /* * fcs_port_api BFA FCS port API */ struct bfa_fcs_lport_s * bfa_fcs_get_base_port(struct bfa_fcs_s *fcs) { return &fcs->fabric.bport; } wwn_t bfa_fcs_lport_get_rport(struct bfa_fcs_lport_s *port, wwn_t wwn, int index, int nrports, bfa_boolean_t bwwn) { struct list_head *qh, *qe; struct bfa_fcs_rport_s *rport = NULL; int i; struct bfa_fcs_s *fcs; if (port == NULL || nrports == 0) return (wwn_t) 0; fcs = port->fcs; bfa_trc(fcs, (u32) nrports); i = 0; qh = &port->rport_q; qe = bfa_q_first(qh); while ((qe != qh) && (i < nrports)) { rport = (struct bfa_fcs_rport_s *) qe; if (bfa_ntoh3b(rport->pid) > 0xFFF000) { qe = bfa_q_next(qe); bfa_trc(fcs, (u32) rport->pwwn); bfa_trc(fcs, rport->pid); bfa_trc(fcs, i); continue; } if (bwwn) { if (!memcmp(&wwn, &rport->pwwn, 8)) break; } else { if (i == index) break; } i++; qe = bfa_q_next(qe); } bfa_trc(fcs, i); if (rport) return rport->pwwn; else return (wwn_t) 0; } void bfa_fcs_lport_get_rports(struct bfa_fcs_lport_s *port, wwn_t rport_wwns[], int *nrports) { struct list_head *qh, *qe; struct bfa_fcs_rport_s *rport = NULL; int i; struct bfa_fcs_s *fcs; if (port == NULL || rport_wwns == NULL || *nrports == 0) return; fcs = port->fcs; bfa_trc(fcs, (u32) *nrports); i = 0; qh = &port->rport_q; qe = bfa_q_first(qh); while ((qe != qh) && (i < *nrports)) { rport = (struct bfa_fcs_rport_s *) qe; if (bfa_ntoh3b(rport->pid) > 0xFFF000) { qe = bfa_q_next(qe); bfa_trc(fcs, (u32) rport->pwwn); bfa_trc(fcs, rport->pid); bfa_trc(fcs, i); continue; } rport_wwns[i] = rport->pwwn; i++; qe = bfa_q_next(qe); } bfa_trc(fcs, i); *nrports = i; } /* * Iterate's through all the rport's in the given port to * determine the maximum operating speed. * * !!!! To be used in TRL Functionality only !!!! */ bfa_port_speed_t bfa_fcs_lport_get_rport_max_speed(bfa_fcs_lport_t *port) { struct list_head *qh, *qe; struct bfa_fcs_rport_s *rport = NULL; struct bfa_fcs_s *fcs; bfa_port_speed_t max_speed = 0; struct bfa_port_attr_s port_attr; bfa_port_speed_t port_speed, rport_speed; bfa_boolean_t trl_enabled = bfa_fcport_is_ratelim(port->fcs->bfa); if (port == NULL) return 0; fcs = port->fcs; /* Get Physical port's current speed */ bfa_fcport_get_attr(port->fcs->bfa, &port_attr); port_speed = port_attr.speed; bfa_trc(fcs, port_speed); qh = &port->rport_q; qe = bfa_q_first(qh); while (qe != qh) { rport = (struct bfa_fcs_rport_s *) qe; if ((bfa_ntoh3b(rport->pid) > 0xFFF000) || (bfa_fcs_rport_get_state(rport) == BFA_RPORT_OFFLINE)) { qe = bfa_q_next(qe); continue; } rport_speed = rport->rpf.rpsc_speed; if ((trl_enabled) && (rport_speed == BFA_PORT_SPEED_UNKNOWN)) { /* Use default ratelim speed setting */ rport_speed = bfa_fcport_get_ratelim_speed(port->fcs->bfa); } if ((rport_speed == BFA_PORT_SPEED_8GBPS) || (rport_speed > port_speed)) { max_speed = rport_speed; break; } else if (rport_speed > max_speed) { max_speed = rport_speed; } qe = bfa_q_next(qe); } bfa_trc(fcs, max_speed); return max_speed; } struct bfa_fcs_lport_s * bfa_fcs_lookup_port(struct bfa_fcs_s *fcs, u16 vf_id, wwn_t lpwwn) { struct bfa_fcs_vport_s *vport; bfa_fcs_vf_t *vf; WARN_ON(fcs == NULL); vf = bfa_fcs_vf_lookup(fcs, vf_id); if (vf == NULL) { bfa_trc(fcs, vf_id); return NULL; } if (!lpwwn || (vf->bport.port_cfg.pwwn == lpwwn)) return &vf->bport; vport = bfa_fcs_fabric_vport_lookup(vf, lpwwn); if (vport) return &vport->lport; return NULL; } /* * API corresponding to NPIV_VPORT_GETINFO. */ void bfa_fcs_lport_get_info(struct bfa_fcs_lport_s *port, struct bfa_lport_info_s *port_info) { bfa_trc(port->fcs, port->fabric->fabric_name); if (port->vport == NULL) { /* * This is a Physical port */ port_info->port_type = BFA_LPORT_TYPE_PHYSICAL; /* * @todo : need to fix the state & reason */ port_info->port_state = 0; port_info->offline_reason = 0; port_info->port_wwn = bfa_fcs_lport_get_pwwn(port); port_info->node_wwn = bfa_fcs_lport_get_nwwn(port); port_info->max_vports_supp = bfa_lps_get_max_vport(port->fcs->bfa); port_info->num_vports_inuse = port->fabric->num_vports; port_info->max_rports_supp = BFA_FCS_MAX_RPORTS_SUPP; port_info->num_rports_inuse = port->num_rports; } else { /* * This is a virtual port */ port_info->port_type = BFA_LPORT_TYPE_VIRTUAL; /* * @todo : need to fix the state & reason */ port_info->port_state = 0; port_info->offline_reason = 0; port_info->port_wwn = bfa_fcs_lport_get_pwwn(port); port_info->node_wwn = bfa_fcs_lport_get_nwwn(port); } } void bfa_fcs_lport_get_stats(struct bfa_fcs_lport_s *fcs_port, struct bfa_lport_stats_s *port_stats) { *port_stats = fcs_port->stats; } void bfa_fcs_lport_clear_stats(struct bfa_fcs_lport_s *fcs_port) { memset(&fcs_port->stats, 0, sizeof(struct bfa_lport_stats_s)); } /* * FCS virtual port state machine */ #define __vport_fcs(__vp) ((__vp)->lport.fcs) #define __vport_pwwn(__vp) ((__vp)->lport.port_cfg.pwwn) #define __vport_nwwn(__vp) ((__vp)->lport.port_cfg.nwwn) #define __vport_bfa(__vp) ((__vp)->lport.fcs->bfa) #define __vport_fcid(__vp) ((__vp)->lport.pid) #define __vport_fabric(__vp) ((__vp)->lport.fabric) #define __vport_vfid(__vp) ((__vp)->lport.fabric->vf_id) #define BFA_FCS_VPORT_MAX_RETRIES 5 /* * Forward declarations */ static void bfa_fcs_vport_do_fdisc(struct bfa_fcs_vport_s *vport); static void bfa_fcs_vport_timeout(void *vport_arg); static void bfa_fcs_vport_do_logo(struct bfa_fcs_vport_s *vport); static void bfa_fcs_vport_free(struct bfa_fcs_vport_s *vport); /* * fcs_vport_sm FCS virtual port state machine */ /* * VPort State Machine events */ enum bfa_fcs_vport_event { BFA_FCS_VPORT_SM_CREATE = 1, /* vport create event */ BFA_FCS_VPORT_SM_DELETE = 2, /* vport delete event */ BFA_FCS_VPORT_SM_START = 3, /* vport start request */ BFA_FCS_VPORT_SM_STOP = 4, /* stop: unsupported */ BFA_FCS_VPORT_SM_ONLINE = 5, /* fabric online */ BFA_FCS_VPORT_SM_OFFLINE = 6, /* fabric offline event */ BFA_FCS_VPORT_SM_FRMSENT = 7, /* fdisc/logo sent events */ BFA_FCS_VPORT_SM_RSP_OK = 8, /* good response */ BFA_FCS_VPORT_SM_RSP_ERROR = 9, /* error/bad response */ BFA_FCS_VPORT_SM_TIMEOUT = 10, /* delay timer event */ BFA_FCS_VPORT_SM_DELCOMP = 11, /* lport delete completion */ BFA_FCS_VPORT_SM_RSP_DUP_WWN = 12, /* Dup wnn error*/ BFA_FCS_VPORT_SM_RSP_FAILED = 13, /* non-retryable failure */ BFA_FCS_VPORT_SM_STOPCOMP = 14, /* vport delete completion */ }; static void bfa_fcs_vport_sm_uninit(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_created(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_offline(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_fdisc(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_fdisc_retry(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_online(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_deleting(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_cleanup(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_logo(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_error(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_stopping(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static void bfa_fcs_vport_sm_logo_for_stop(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event); static struct bfa_sm_table_s vport_sm_table[] = { {BFA_SM(bfa_fcs_vport_sm_uninit), BFA_FCS_VPORT_UNINIT}, {BFA_SM(bfa_fcs_vport_sm_created), BFA_FCS_VPORT_CREATED}, {BFA_SM(bfa_fcs_vport_sm_offline), BFA_FCS_VPORT_OFFLINE}, {BFA_SM(bfa_fcs_vport_sm_fdisc), BFA_FCS_VPORT_FDISC}, {BFA_SM(bfa_fcs_vport_sm_fdisc_retry), BFA_FCS_VPORT_FDISC_RETRY}, {BFA_SM(bfa_fcs_vport_sm_online), BFA_FCS_VPORT_ONLINE}, {BFA_SM(bfa_fcs_vport_sm_deleting), BFA_FCS_VPORT_DELETING}, {BFA_SM(bfa_fcs_vport_sm_cleanup), BFA_FCS_VPORT_CLEANUP}, {BFA_SM(bfa_fcs_vport_sm_logo), BFA_FCS_VPORT_LOGO}, {BFA_SM(bfa_fcs_vport_sm_error), BFA_FCS_VPORT_ERROR} }; /* * Beginning state. */ static void bfa_fcs_vport_sm_uninit(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_CREATE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_created); bfa_fcs_fabric_addvport(__vport_fabric(vport), vport); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * Created state - a start event is required to start up the state machine. */ static void bfa_fcs_vport_sm_created(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_START: if (bfa_sm_cmp_state(__vport_fabric(vport), bfa_fcs_fabric_sm_online) && bfa_fcs_fabric_npiv_capable(__vport_fabric(vport))) { bfa_sm_set_state(vport, bfa_fcs_vport_sm_fdisc); bfa_fcs_vport_do_fdisc(vport); } else { /* * Fabric is offline or not NPIV capable, stay in * offline state. */ vport->vport_stats.fab_no_npiv++; bfa_sm_set_state(vport, bfa_fcs_vport_sm_offline); } break; case BFA_FCS_VPORT_SM_DELETE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); bfa_fcs_lport_delete(&vport->lport); break; case BFA_FCS_VPORT_SM_ONLINE: case BFA_FCS_VPORT_SM_OFFLINE: /* * Ignore ONLINE/OFFLINE events from fabric * till vport is started. */ break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * Offline state - awaiting ONLINE event from fabric SM. */ static void bfa_fcs_vport_sm_offline(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELETE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); bfa_fcs_lport_delete(&vport->lport); break; case BFA_FCS_VPORT_SM_ONLINE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_fdisc); vport->fdisc_retries = 0; bfa_fcs_vport_do_fdisc(vport); break; case BFA_FCS_VPORT_SM_STOP: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); bfa_sm_send_event(&vport->lport, BFA_FCS_PORT_SM_STOP); break; case BFA_FCS_VPORT_SM_OFFLINE: /* * This can happen if the vport couldn't be initialzied * due the fact that the npiv was not enabled on the switch. * In that case we will put the vport in offline state. * However, the link can go down and cause the this event to * be sent when we are already offline. Ignore it. */ break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * FDISC is sent and awaiting reply from fabric. */ static void bfa_fcs_vport_sm_fdisc(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELETE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); bfa_sm_send_event(vport->lps, BFA_LPS_SM_OFFLINE); bfa_fcs_lport_delete(&vport->lport); break; case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_offline); bfa_sm_send_event(vport->lps, BFA_LPS_SM_OFFLINE); break; case BFA_FCS_VPORT_SM_RSP_OK: bfa_sm_set_state(vport, bfa_fcs_vport_sm_online); bfa_fcs_lport_online(&vport->lport); break; case BFA_FCS_VPORT_SM_RSP_ERROR: bfa_sm_set_state(vport, bfa_fcs_vport_sm_fdisc_retry); bfa_timer_start(__vport_bfa(vport), &vport->timer, bfa_fcs_vport_timeout, vport, BFA_FCS_RETRY_TIMEOUT); break; case BFA_FCS_VPORT_SM_RSP_FAILED: bfa_sm_set_state(vport, bfa_fcs_vport_sm_offline); break; case BFA_FCS_VPORT_SM_RSP_DUP_WWN: bfa_sm_set_state(vport, bfa_fcs_vport_sm_error); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * FDISC attempt failed - a timer is active to retry FDISC. */ static void bfa_fcs_vport_sm_fdisc_retry(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELETE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); bfa_timer_stop(&vport->timer); bfa_fcs_lport_delete(&vport->lport); break; case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_offline); bfa_timer_stop(&vport->timer); break; case BFA_FCS_VPORT_SM_TIMEOUT: bfa_sm_set_state(vport, bfa_fcs_vport_sm_fdisc); vport->vport_stats.fdisc_retries++; vport->fdisc_retries++; bfa_fcs_vport_do_fdisc(vport); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * Vport is online (FDISC is complete). */ static void bfa_fcs_vport_sm_online(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELETE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_deleting); bfa_fcs_lport_delete(&vport->lport); break; case BFA_FCS_VPORT_SM_STOP: bfa_sm_set_state(vport, bfa_fcs_vport_sm_stopping); bfa_sm_send_event(&vport->lport, BFA_FCS_PORT_SM_STOP); break; case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_offline); bfa_sm_send_event(vport->lps, BFA_LPS_SM_OFFLINE); bfa_fcs_lport_offline(&vport->lport); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * Vport is being stopped - awaiting lport stop completion to send * LOGO to fabric. */ static void bfa_fcs_vport_sm_stopping(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_STOPCOMP: bfa_sm_set_state(vport, bfa_fcs_vport_sm_logo_for_stop); bfa_fcs_vport_do_logo(vport); break; case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * Vport is being deleted - awaiting lport delete completion to send * LOGO to fabric. */ static void bfa_fcs_vport_sm_deleting(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELETE: break; case BFA_FCS_VPORT_SM_DELCOMP: bfa_sm_set_state(vport, bfa_fcs_vport_sm_logo); bfa_fcs_vport_do_logo(vport); break; case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * Error State. * This state will be set when the Vport Creation fails due * to errors like Dup WWN. In this state only operation allowed * is a Vport Delete. */ static void bfa_fcs_vport_sm_error(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELETE: bfa_sm_set_state(vport, bfa_fcs_vport_sm_cleanup); bfa_fcs_lport_delete(&vport->lport); break; default: bfa_trc(__vport_fcs(vport), event); } } /* * Lport cleanup is in progress since vport is being deleted. Fabric is * offline, so no LOGO is needed to complete vport deletion. */ static void bfa_fcs_vport_sm_cleanup(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_DELCOMP: bfa_sm_set_state(vport, bfa_fcs_vport_sm_uninit); bfa_fcs_vport_free(vport); break; case BFA_FCS_VPORT_SM_STOPCOMP: bfa_sm_set_state(vport, bfa_fcs_vport_sm_created); break; case BFA_FCS_VPORT_SM_DELETE: break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * LOGO is sent to fabric. Vport stop is in progress. Lport stop cleanup * is done. */ static void bfa_fcs_vport_sm_logo_for_stop(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_send_event(vport->lps, BFA_LPS_SM_OFFLINE); /* * !!! fall through !!! */ case BFA_FCS_VPORT_SM_RSP_OK: case BFA_FCS_VPORT_SM_RSP_ERROR: bfa_sm_set_state(vport, bfa_fcs_vport_sm_created); break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * LOGO is sent to fabric. Vport delete is in progress. Lport delete cleanup * is done. */ static void bfa_fcs_vport_sm_logo(struct bfa_fcs_vport_s *vport, enum bfa_fcs_vport_event event) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), event); switch (event) { case BFA_FCS_VPORT_SM_OFFLINE: bfa_sm_send_event(vport->lps, BFA_LPS_SM_OFFLINE); /* * !!! fall through !!! */ case BFA_FCS_VPORT_SM_RSP_OK: case BFA_FCS_VPORT_SM_RSP_ERROR: bfa_sm_set_state(vport, bfa_fcs_vport_sm_uninit); bfa_fcs_vport_free(vport); break; case BFA_FCS_VPORT_SM_DELETE: break; default: bfa_sm_fault(__vport_fcs(vport), event); } } /* * fcs_vport_private FCS virtual port private functions */ /* * This routine will be called to send a FDISC command. */ static void bfa_fcs_vport_do_fdisc(struct bfa_fcs_vport_s *vport) { bfa_lps_fdisc(vport->lps, vport, bfa_fcport_get_maxfrsize(__vport_bfa(vport)), __vport_pwwn(vport), __vport_nwwn(vport)); vport->vport_stats.fdisc_sent++; } static void bfa_fcs_vport_fdisc_rejected(struct bfa_fcs_vport_s *vport) { u8 lsrjt_rsn = vport->lps->lsrjt_rsn; u8 lsrjt_expl = vport->lps->lsrjt_expl; bfa_trc(__vport_fcs(vport), lsrjt_rsn); bfa_trc(__vport_fcs(vport), lsrjt_expl); /* For certain reason codes, we don't want to retry. */ switch (vport->lps->lsrjt_expl) { case FC_LS_RJT_EXP_INV_PORT_NAME: /* by brocade */ case FC_LS_RJT_EXP_INVALID_NPORT_ID: /* by Cisco */ if (vport->fdisc_retries < BFA_FCS_VPORT_MAX_RETRIES) bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_ERROR); else bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_DUP_WWN); break; case FC_LS_RJT_EXP_INSUFF_RES: /* * This means max logins per port/switch setting on the * switch was exceeded. */ if (vport->fdisc_retries < BFA_FCS_VPORT_MAX_RETRIES) bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_ERROR); else bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_FAILED); break; default: bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_ERROR); } } /* * Called to send a logout to the fabric. Used when a V-Port is * deleted/stopped. */ static void bfa_fcs_vport_do_logo(struct bfa_fcs_vport_s *vport) { bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); vport->vport_stats.logo_sent++; bfa_lps_fdisclogo(vport->lps); } /* * This routine will be called by bfa_timer on timer timeouts. * * param[in] vport - pointer to bfa_fcs_vport_t. * param[out] vport_status - pointer to return vport status in * * return * void * * Special Considerations: * * note */ static void bfa_fcs_vport_timeout(void *vport_arg) { struct bfa_fcs_vport_s *vport = (struct bfa_fcs_vport_s *) vport_arg; vport->vport_stats.fdisc_timeouts++; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_TIMEOUT); } static void bfa_fcs_vport_free(struct bfa_fcs_vport_s *vport) { struct bfad_vport_s *vport_drv = (struct bfad_vport_s *)vport->vport_drv; bfa_fcs_fabric_delvport(__vport_fabric(vport), vport); if (vport_drv->comp_del) complete(vport_drv->comp_del); bfa_lps_delete(vport->lps); } /* * fcs_vport_public FCS virtual port public interfaces */ /* * Online notification from fabric SM. */ void bfa_fcs_vport_online(struct bfa_fcs_vport_s *vport) { vport->vport_stats.fab_online++; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_ONLINE); } /* * Offline notification from fabric SM. */ void bfa_fcs_vport_offline(struct bfa_fcs_vport_s *vport) { vport->vport_stats.fab_offline++; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_OFFLINE); } /* * Cleanup notification from fabric SM on link timer expiry. */ void bfa_fcs_vport_cleanup(struct bfa_fcs_vport_s *vport) { vport->vport_stats.fab_cleanup++; } /* * delete notification from fabric SM. To be invoked from within FCS. */ void bfa_fcs_vport_fcs_delete(struct bfa_fcs_vport_s *vport) { bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_DELETE); } /* * Stop completion callback from associated lport */ void bfa_fcs_vport_stop_comp(struct bfa_fcs_vport_s *vport) { bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_STOPCOMP); } /* * Delete completion callback from associated lport */ void bfa_fcs_vport_delete_comp(struct bfa_fcs_vport_s *vport) { bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_DELCOMP); } /* * fcs_vport_api Virtual port API */ /* * Use this function to instantiate a new FCS vport object. This * function will not trigger any HW initialization process (which will be * done in vport_start() call) * * param[in] vport - pointer to bfa_fcs_vport_t. This space * needs to be allocated by the driver. * param[in] fcs - FCS instance * param[in] vport_cfg - vport configuration * param[in] vf_id - VF_ID if vport is created within a VF. * FC_VF_ID_NULL to specify base fabric. * param[in] vport_drv - Opaque handle back to the driver's vport * structure * * retval BFA_STATUS_OK - on success. * retval BFA_STATUS_FAILED - on failure. */ bfa_status_t bfa_fcs_vport_create(struct bfa_fcs_vport_s *vport, struct bfa_fcs_s *fcs, u16 vf_id, struct bfa_lport_cfg_s *vport_cfg, struct bfad_vport_s *vport_drv) { if (vport_cfg->pwwn == 0) return BFA_STATUS_INVALID_WWN; if (bfa_fcs_lport_get_pwwn(&fcs->fabric.bport) == vport_cfg->pwwn) return BFA_STATUS_VPORT_WWN_BP; if (bfa_fcs_vport_lookup(fcs, vf_id, vport_cfg->pwwn) != NULL) return BFA_STATUS_VPORT_EXISTS; if (fcs->fabric.num_vports == bfa_lps_get_max_vport(fcs->bfa)) return BFA_STATUS_VPORT_MAX; vport->lps = bfa_lps_alloc(fcs->bfa); if (!vport->lps) return BFA_STATUS_VPORT_MAX; vport->vport_drv = vport_drv; vport_cfg->preboot_vp = BFA_FALSE; bfa_sm_set_state(vport, bfa_fcs_vport_sm_uninit); bfa_fcs_lport_attach(&vport->lport, fcs, vf_id, vport); bfa_fcs_lport_init(&vport->lport, vport_cfg); bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_CREATE); return BFA_STATUS_OK; } /* * Use this function to instantiate a new FCS PBC vport object. This * function will not trigger any HW initialization process (which will be * done in vport_start() call) * * param[in] vport - pointer to bfa_fcs_vport_t. This space * needs to be allocated by the driver. * param[in] fcs - FCS instance * param[in] vport_cfg - vport configuration * param[in] vf_id - VF_ID if vport is created within a VF. * FC_VF_ID_NULL to specify base fabric. * param[in] vport_drv - Opaque handle back to the driver's vport * structure * * retval BFA_STATUS_OK - on success. * retval BFA_STATUS_FAILED - on failure. */ bfa_status_t bfa_fcs_pbc_vport_create(struct bfa_fcs_vport_s *vport, struct bfa_fcs_s *fcs, u16 vf_id, struct bfa_lport_cfg_s *vport_cfg, struct bfad_vport_s *vport_drv) { bfa_status_t rc; rc = bfa_fcs_vport_create(vport, fcs, vf_id, vport_cfg, vport_drv); vport->lport.port_cfg.preboot_vp = BFA_TRUE; return rc; } /* * Use this function to findout if this is a pbc vport or not. * * @param[in] vport - pointer to bfa_fcs_vport_t. * * @returns None */ bfa_boolean_t bfa_fcs_is_pbc_vport(struct bfa_fcs_vport_s *vport) { if (vport && (vport->lport.port_cfg.preboot_vp == BFA_TRUE)) return BFA_TRUE; else return BFA_FALSE; } /* * Use this function initialize the vport. * * @param[in] vport - pointer to bfa_fcs_vport_t. * * @returns None */ bfa_status_t bfa_fcs_vport_start(struct bfa_fcs_vport_s *vport) { bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_START); return BFA_STATUS_OK; } /* * Use this function quiese the vport object. This function will return * immediately, when the vport is actually stopped, the * bfa_drv_vport_stop_cb() will be called. * * param[in] vport - pointer to bfa_fcs_vport_t. * * return None */ bfa_status_t bfa_fcs_vport_stop(struct bfa_fcs_vport_s *vport) { bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_STOP); return BFA_STATUS_OK; } /* * Use this function to delete a vport object. Fabric object should * be stopped before this function call. * * !!!!!!! Donot invoke this from within FCS !!!!!!! * * param[in] vport - pointer to bfa_fcs_vport_t. * * return None */ bfa_status_t bfa_fcs_vport_delete(struct bfa_fcs_vport_s *vport) { if (vport->lport.port_cfg.preboot_vp) return BFA_STATUS_PBC; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_DELETE); return BFA_STATUS_OK; } /* * Use this function to get vport's current status info. * * param[in] vport pointer to bfa_fcs_vport_t. * param[out] attr pointer to return vport attributes * * return None */ void bfa_fcs_vport_get_attr(struct bfa_fcs_vport_s *vport, struct bfa_vport_attr_s *attr) { if (vport == NULL || attr == NULL) return; memset(attr, 0, sizeof(struct bfa_vport_attr_s)); bfa_fcs_lport_get_attr(&vport->lport, &attr->port_attr); attr->vport_state = bfa_sm_to_state(vport_sm_table, vport->sm); } /* * Lookup a virtual port. Excludes base port from lookup. */ struct bfa_fcs_vport_s * bfa_fcs_vport_lookup(struct bfa_fcs_s *fcs, u16 vf_id, wwn_t vpwwn) { struct bfa_fcs_vport_s *vport; struct bfa_fcs_fabric_s *fabric; bfa_trc(fcs, vf_id); bfa_trc(fcs, vpwwn); fabric = bfa_fcs_vf_lookup(fcs, vf_id); if (!fabric) { bfa_trc(fcs, vf_id); return NULL; } vport = bfa_fcs_fabric_vport_lookup(fabric, vpwwn); return vport; } /* * FDISC Response */ void bfa_cb_lps_fdisc_comp(void *bfad, void *uarg, bfa_status_t status) { struct bfa_fcs_vport_s *vport = uarg; bfa_trc(__vport_fcs(vport), __vport_pwwn(vport)); bfa_trc(__vport_fcs(vport), status); switch (status) { case BFA_STATUS_OK: /* * Initialize the V-Port fields */ __vport_fcid(vport) = vport->lps->lp_pid; vport->vport_stats.fdisc_accepts++; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_OK); break; case BFA_STATUS_INVALID_MAC: /* Only for CNA */ vport->vport_stats.fdisc_acc_bad++; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_ERROR); break; case BFA_STATUS_EPROTOCOL: switch (vport->lps->ext_status) { case BFA_EPROTO_BAD_ACCEPT: vport->vport_stats.fdisc_acc_bad++; break; case BFA_EPROTO_UNKNOWN_RSP: vport->vport_stats.fdisc_unknown_rsp++; break; default: break; } bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_ERROR); break; case BFA_STATUS_FABRIC_RJT: vport->vport_stats.fdisc_rejects++; bfa_fcs_vport_fdisc_rejected(vport); break; default: vport->vport_stats.fdisc_rsp_err++; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_ERROR); } } /* * LOGO response */ void bfa_cb_lps_fdisclogo_comp(void *bfad, void *uarg) { struct bfa_fcs_vport_s *vport = uarg; bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_RSP_OK); } /* * Received clear virtual link */ void bfa_cb_lps_cvl_event(void *bfad, void *uarg) { struct bfa_fcs_vport_s *vport = uarg; /* Send an Offline followed by an ONLINE */ bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_OFFLINE); bfa_sm_send_event(vport, BFA_FCS_VPORT_SM_ONLINE); }