/* * Copyright (c) 2005-2014 Brocade Communications Systems, Inc. * Copyright (c) 2014- QLogic Corporation. * All rights reserved * www.qlogic.com * * Linux driver for QLogic BR-series 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 "bfad_im.h" #include "bfa_plog.h" #include "bfa_cs.h" #include "bfa_modules.h" BFA_TRC_FILE(HAL, FCXP); /* * LPS related definitions */ #define BFA_LPS_MIN_LPORTS (1) #define BFA_LPS_MAX_LPORTS (256) /* * Maximum Vports supported per physical port or vf. */ #define BFA_LPS_MAX_VPORTS_SUPP_CB 255 #define BFA_LPS_MAX_VPORTS_SUPP_CT 190 /* * FC PORT related definitions */ /* * The port is considered disabled if corresponding physical port or IOC are * disabled explicitly */ #define BFA_PORT_IS_DISABLED(bfa) \ ((bfa_fcport_is_disabled(bfa) == BFA_TRUE) || \ (bfa_ioc_is_disabled(&bfa->ioc) == BFA_TRUE)) /* * BFA port state machine events */ enum bfa_fcport_sm_event { BFA_FCPORT_SM_START = 1, /* start port state machine */ BFA_FCPORT_SM_STOP = 2, /* stop port state machine */ BFA_FCPORT_SM_ENABLE = 3, /* enable port */ BFA_FCPORT_SM_DISABLE = 4, /* disable port state machine */ BFA_FCPORT_SM_FWRSP = 5, /* firmware enable/disable rsp */ BFA_FCPORT_SM_LINKUP = 6, /* firmware linkup event */ BFA_FCPORT_SM_LINKDOWN = 7, /* firmware linkup down */ BFA_FCPORT_SM_QRESUME = 8, /* CQ space available */ BFA_FCPORT_SM_HWFAIL = 9, /* IOC h/w failure */ BFA_FCPORT_SM_DPORTENABLE = 10, /* enable dport */ BFA_FCPORT_SM_DPORTDISABLE = 11,/* disable dport */ BFA_FCPORT_SM_FAA_MISCONFIG = 12, /* FAA misconfiguratin */ BFA_FCPORT_SM_DDPORTENABLE = 13, /* enable ddport */ BFA_FCPORT_SM_DDPORTDISABLE = 14, /* disable ddport */ }; /* * BFA port link notification state machine events */ enum bfa_fcport_ln_sm_event { BFA_FCPORT_LN_SM_LINKUP = 1, /* linkup event */ BFA_FCPORT_LN_SM_LINKDOWN = 2, /* linkdown event */ BFA_FCPORT_LN_SM_NOTIFICATION = 3 /* done notification */ }; /* * RPORT related definitions */ #define bfa_rport_offline_cb(__rp) do { \ if ((__rp)->bfa->fcs) \ bfa_cb_rport_offline((__rp)->rport_drv); \ else { \ bfa_cb_queue((__rp)->bfa, &(__rp)->hcb_qe, \ __bfa_cb_rport_offline, (__rp)); \ } \ } while (0) #define bfa_rport_online_cb(__rp) do { \ if ((__rp)->bfa->fcs) \ bfa_cb_rport_online((__rp)->rport_drv); \ else { \ bfa_cb_queue((__rp)->bfa, &(__rp)->hcb_qe, \ __bfa_cb_rport_online, (__rp)); \ } \ } while (0) /* * forward declarations FCXP related functions */ static void __bfa_fcxp_send_cbfn(void *cbarg, bfa_boolean_t complete); static void hal_fcxp_rx_plog(struct bfa_s *bfa, struct bfa_fcxp_s *fcxp, struct bfi_fcxp_send_rsp_s *fcxp_rsp); static void hal_fcxp_tx_plog(struct bfa_s *bfa, u32 reqlen, struct bfa_fcxp_s *fcxp, struct fchs_s *fchs); static void bfa_fcxp_qresume(void *cbarg); static void bfa_fcxp_queue(struct bfa_fcxp_s *fcxp, struct bfi_fcxp_send_req_s *send_req); /* * forward declarations for LPS functions */ static void bfa_lps_login_rsp(struct bfa_s *bfa, struct bfi_lps_login_rsp_s *rsp); static void bfa_lps_no_res(struct bfa_lps_s *first_lps, u8 count); static void bfa_lps_logout_rsp(struct bfa_s *bfa, struct bfi_lps_logout_rsp_s *rsp); static void bfa_lps_reqq_resume(void *lps_arg); static void bfa_lps_free(struct bfa_lps_s *lps); static void bfa_lps_send_login(struct bfa_lps_s *lps); static void bfa_lps_send_logout(struct bfa_lps_s *lps); static void bfa_lps_send_set_n2n_pid(struct bfa_lps_s *lps); static void bfa_lps_login_comp(struct bfa_lps_s *lps); static void bfa_lps_logout_comp(struct bfa_lps_s *lps); static void bfa_lps_cvl_event(struct bfa_lps_s *lps); /* * forward declaration for LPS state machine */ static void bfa_lps_sm_init(struct bfa_lps_s *lps, enum bfa_lps_event event); static void bfa_lps_sm_login(struct bfa_lps_s *lps, enum bfa_lps_event event); static void bfa_lps_sm_loginwait(struct bfa_lps_s *lps, enum bfa_lps_event event); static void bfa_lps_sm_online(struct bfa_lps_s *lps, enum bfa_lps_event event); static void bfa_lps_sm_online_n2n_pid_wait(struct bfa_lps_s *lps, enum bfa_lps_event event); static void bfa_lps_sm_logout(struct bfa_lps_s *lps, enum bfa_lps_event event); static void bfa_lps_sm_logowait(struct bfa_lps_s *lps, enum bfa_lps_event event); /* * forward declaration for FC Port functions */ static bfa_boolean_t bfa_fcport_send_enable(struct bfa_fcport_s *fcport); static bfa_boolean_t bfa_fcport_send_disable(struct bfa_fcport_s *fcport); static void bfa_fcport_update_linkinfo(struct bfa_fcport_s *fcport); static void bfa_fcport_reset_linkinfo(struct bfa_fcport_s *fcport); static void bfa_fcport_set_wwns(struct bfa_fcport_s *fcport); static void __bfa_cb_fcport_event(void *cbarg, bfa_boolean_t complete); static void bfa_fcport_scn(struct bfa_fcport_s *fcport, enum bfa_port_linkstate event, bfa_boolean_t trunk); static void bfa_fcport_queue_cb(struct bfa_fcport_ln_s *ln, enum bfa_port_linkstate event); static void __bfa_cb_fcport_stats_clr(void *cbarg, bfa_boolean_t complete); static void bfa_fcport_stats_get_timeout(void *cbarg); static void bfa_fcport_stats_clr_timeout(void *cbarg); static void bfa_trunk_iocdisable(struct bfa_s *bfa); /* * forward declaration for FC PORT state machine */ static void bfa_fcport_sm_uninit(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_enabling_qwait(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_enabling(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_linkdown(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_linkup(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_disabling(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_disabling_qwait(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_toggling_qwait(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_disabled(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_stopped(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_iocdown(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_iocfail(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_dport(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_ddport(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_sm_faa_misconfig(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event); static void bfa_fcport_ln_sm_dn(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static void bfa_fcport_ln_sm_dn_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static void bfa_fcport_ln_sm_dn_up_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static void bfa_fcport_ln_sm_up(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static void bfa_fcport_ln_sm_up_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static void bfa_fcport_ln_sm_up_dn_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static void bfa_fcport_ln_sm_up_dn_up_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event); static struct bfa_sm_table_s hal_port_sm_table[] = { {BFA_SM(bfa_fcport_sm_uninit), BFA_PORT_ST_UNINIT}, {BFA_SM(bfa_fcport_sm_enabling_qwait), BFA_PORT_ST_ENABLING_QWAIT}, {BFA_SM(bfa_fcport_sm_enabling), BFA_PORT_ST_ENABLING}, {BFA_SM(bfa_fcport_sm_linkdown), BFA_PORT_ST_LINKDOWN}, {BFA_SM(bfa_fcport_sm_linkup), BFA_PORT_ST_LINKUP}, {BFA_SM(bfa_fcport_sm_disabling_qwait), BFA_PORT_ST_DISABLING_QWAIT}, {BFA_SM(bfa_fcport_sm_toggling_qwait), BFA_PORT_ST_TOGGLING_QWAIT}, {BFA_SM(bfa_fcport_sm_disabling), BFA_PORT_ST_DISABLING}, {BFA_SM(bfa_fcport_sm_disabled), BFA_PORT_ST_DISABLED}, {BFA_SM(bfa_fcport_sm_stopped), BFA_PORT_ST_STOPPED}, {BFA_SM(bfa_fcport_sm_iocdown), BFA_PORT_ST_IOCDOWN}, {BFA_SM(bfa_fcport_sm_iocfail), BFA_PORT_ST_IOCDOWN}, {BFA_SM(bfa_fcport_sm_dport), BFA_PORT_ST_DPORT}, {BFA_SM(bfa_fcport_sm_ddport), BFA_PORT_ST_DDPORT}, {BFA_SM(bfa_fcport_sm_faa_misconfig), BFA_PORT_ST_FAA_MISCONFIG}, }; /* * forward declaration for RPORT related functions */ static struct bfa_rport_s *bfa_rport_alloc(struct bfa_rport_mod_s *rp_mod); static void bfa_rport_free(struct bfa_rport_s *rport); static bfa_boolean_t bfa_rport_send_fwcreate(struct bfa_rport_s *rp); static bfa_boolean_t bfa_rport_send_fwdelete(struct bfa_rport_s *rp); static bfa_boolean_t bfa_rport_send_fwspeed(struct bfa_rport_s *rp); static void __bfa_cb_rport_online(void *cbarg, bfa_boolean_t complete); static void __bfa_cb_rport_offline(void *cbarg, bfa_boolean_t complete); /* * forward declaration for RPORT state machine */ static void bfa_rport_sm_uninit(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_created(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_fwcreate(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_online(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_fwdelete(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_offline(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_deleting(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_offline_pending(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_delete_pending(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_iocdisable(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_fwcreate_qfull(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_fwdelete_qfull(struct bfa_rport_s *rp, enum bfa_rport_event event); static void bfa_rport_sm_deleting_qfull(struct bfa_rport_s *rp, enum bfa_rport_event event); /* * PLOG related definitions */ static int plkd_validate_logrec(struct bfa_plog_rec_s *pl_rec) { if ((pl_rec->log_type != BFA_PL_LOG_TYPE_INT) && (pl_rec->log_type != BFA_PL_LOG_TYPE_STRING)) return 1; if ((pl_rec->log_type != BFA_PL_LOG_TYPE_INT) && (pl_rec->log_num_ints > BFA_PL_INT_LOG_SZ)) return 1; return 0; } static u64 bfa_get_log_time(void) { u64 system_time = 0; struct timeval tv; do_gettimeofday(&tv); /* We are interested in seconds only. */ system_time = tv.tv_sec; return system_time; } static void bfa_plog_add(struct bfa_plog_s *plog, struct bfa_plog_rec_s *pl_rec) { u16 tail; struct bfa_plog_rec_s *pl_recp; if (plog->plog_enabled == 0) return; if (plkd_validate_logrec(pl_rec)) { WARN_ON(1); return; } tail = plog->tail; pl_recp = &(plog->plog_recs[tail]); memcpy(pl_recp, pl_rec, sizeof(struct bfa_plog_rec_s)); pl_recp->tv = bfa_get_log_time(); BFA_PL_LOG_REC_INCR(plog->tail); if (plog->head == plog->tail) BFA_PL_LOG_REC_INCR(plog->head); } void bfa_plog_init(struct bfa_plog_s *plog) { memset((char *)plog, 0, sizeof(struct bfa_plog_s)); memcpy(plog->plog_sig, BFA_PL_SIG_STR, BFA_PL_SIG_LEN); plog->head = plog->tail = 0; plog->plog_enabled = 1; } void bfa_plog_str(struct bfa_plog_s *plog, enum bfa_plog_mid mid, enum bfa_plog_eid event, u16 misc, char *log_str) { struct bfa_plog_rec_s lp; if (plog->plog_enabled) { memset(&lp, 0, sizeof(struct bfa_plog_rec_s)); lp.mid = mid; lp.eid = event; lp.log_type = BFA_PL_LOG_TYPE_STRING; lp.misc = misc; strlcpy(lp.log_entry.string_log, log_str, BFA_PL_STRING_LOG_SZ); lp.log_entry.string_log[BFA_PL_STRING_LOG_SZ - 1] = '\0'; bfa_plog_add(plog, &lp); } } void bfa_plog_intarr(struct bfa_plog_s *plog, enum bfa_plog_mid mid, enum bfa_plog_eid event, u16 misc, u32 *intarr, u32 num_ints) { struct bfa_plog_rec_s lp; u32 i; if (num_ints > BFA_PL_INT_LOG_SZ) num_ints = BFA_PL_INT_LOG_SZ; if (plog->plog_enabled) { memset(&lp, 0, sizeof(struct bfa_plog_rec_s)); lp.mid = mid; lp.eid = event; lp.log_type = BFA_PL_LOG_TYPE_INT; lp.misc = misc; for (i = 0; i < num_ints; i++) lp.log_entry.int_log[i] = intarr[i]; lp.log_num_ints = (u8) num_ints; bfa_plog_add(plog, &lp); } } void bfa_plog_fchdr(struct bfa_plog_s *plog, enum bfa_plog_mid mid, enum bfa_plog_eid event, u16 misc, struct fchs_s *fchdr) { struct bfa_plog_rec_s lp; u32 *tmp_int = (u32 *) fchdr; u32 ints[BFA_PL_INT_LOG_SZ]; if (plog->plog_enabled) { memset(&lp, 0, sizeof(struct bfa_plog_rec_s)); ints[0] = tmp_int[0]; ints[1] = tmp_int[1]; ints[2] = tmp_int[4]; bfa_plog_intarr(plog, mid, event, misc, ints, 3); } } void bfa_plog_fchdr_and_pl(struct bfa_plog_s *plog, enum bfa_plog_mid mid, enum bfa_plog_eid event, u16 misc, struct fchs_s *fchdr, u32 pld_w0) { struct bfa_plog_rec_s lp; u32 *tmp_int = (u32 *) fchdr; u32 ints[BFA_PL_INT_LOG_SZ]; if (plog->plog_enabled) { memset(&lp, 0, sizeof(struct bfa_plog_rec_s)); ints[0] = tmp_int[0]; ints[1] = tmp_int[1]; ints[2] = tmp_int[4]; ints[3] = pld_w0; bfa_plog_intarr(plog, mid, event, misc, ints, 4); } } /* * fcxp_pvt BFA FCXP private functions */ static void claim_fcxps_mem(struct bfa_fcxp_mod_s *mod) { u16 i; struct bfa_fcxp_s *fcxp; fcxp = (struct bfa_fcxp_s *) bfa_mem_kva_curp(mod); memset(fcxp, 0, sizeof(struct bfa_fcxp_s) * mod->num_fcxps); INIT_LIST_HEAD(&mod->fcxp_req_free_q); INIT_LIST_HEAD(&mod->fcxp_rsp_free_q); INIT_LIST_HEAD(&mod->fcxp_active_q); INIT_LIST_HEAD(&mod->fcxp_req_unused_q); INIT_LIST_HEAD(&mod->fcxp_rsp_unused_q); mod->fcxp_list = fcxp; for (i = 0; i < mod->num_fcxps; i++) { fcxp->fcxp_mod = mod; fcxp->fcxp_tag = i; if (i < (mod->num_fcxps / 2)) { list_add_tail(&fcxp->qe, &mod->fcxp_req_free_q); fcxp->req_rsp = BFA_TRUE; } else { list_add_tail(&fcxp->qe, &mod->fcxp_rsp_free_q); fcxp->req_rsp = BFA_FALSE; } bfa_reqq_winit(&fcxp->reqq_wqe, bfa_fcxp_qresume, fcxp); fcxp->reqq_waiting = BFA_FALSE; fcxp = fcxp + 1; } bfa_mem_kva_curp(mod) = (void *)fcxp; } void bfa_fcxp_meminfo(struct bfa_iocfc_cfg_s *cfg, struct bfa_meminfo_s *minfo, struct bfa_s *bfa) { struct bfa_fcxp_mod_s *fcxp_mod = BFA_FCXP_MOD(bfa); struct bfa_mem_kva_s *fcxp_kva = BFA_MEM_FCXP_KVA(bfa); struct bfa_mem_dma_s *seg_ptr; u16 nsegs, idx, per_seg_fcxp; u16 num_fcxps = cfg->fwcfg.num_fcxp_reqs; u32 per_fcxp_sz; if (num_fcxps == 0) return; if (cfg->drvcfg.min_cfg) per_fcxp_sz = 2 * BFA_FCXP_MAX_IBUF_SZ; else per_fcxp_sz = BFA_FCXP_MAX_IBUF_SZ + BFA_FCXP_MAX_LBUF_SZ; /* dma memory */ nsegs = BFI_MEM_DMA_NSEGS(num_fcxps, per_fcxp_sz); per_seg_fcxp = BFI_MEM_NREQS_SEG(per_fcxp_sz); bfa_mem_dma_seg_iter(fcxp_mod, seg_ptr, nsegs, idx) { if (num_fcxps >= per_seg_fcxp) { num_fcxps -= per_seg_fcxp; bfa_mem_dma_setup(minfo, seg_ptr, per_seg_fcxp * per_fcxp_sz); } else bfa_mem_dma_setup(minfo, seg_ptr, num_fcxps * per_fcxp_sz); } /* kva memory */ bfa_mem_kva_setup(minfo, fcxp_kva, cfg->fwcfg.num_fcxp_reqs * sizeof(struct bfa_fcxp_s)); } void bfa_fcxp_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); mod->bfa = bfa; mod->num_fcxps = cfg->fwcfg.num_fcxp_reqs; /* * Initialize FCXP request and response payload sizes. */ mod->req_pld_sz = mod->rsp_pld_sz = BFA_FCXP_MAX_IBUF_SZ; if (!cfg->drvcfg.min_cfg) mod->rsp_pld_sz = BFA_FCXP_MAX_LBUF_SZ; INIT_LIST_HEAD(&mod->req_wait_q); INIT_LIST_HEAD(&mod->rsp_wait_q); claim_fcxps_mem(mod); } void bfa_fcxp_iocdisable(struct bfa_s *bfa) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); struct bfa_fcxp_s *fcxp; struct list_head *qe, *qen; /* Enqueue unused fcxp resources to free_q */ list_splice_tail_init(&mod->fcxp_req_unused_q, &mod->fcxp_req_free_q); list_splice_tail_init(&mod->fcxp_rsp_unused_q, &mod->fcxp_rsp_free_q); list_for_each_safe(qe, qen, &mod->fcxp_active_q) { fcxp = (struct bfa_fcxp_s *) qe; if (fcxp->caller == NULL) { fcxp->send_cbfn(fcxp->caller, fcxp, fcxp->send_cbarg, BFA_STATUS_IOC_FAILURE, 0, 0, NULL); bfa_fcxp_free(fcxp); } else { fcxp->rsp_status = BFA_STATUS_IOC_FAILURE; bfa_cb_queue(bfa, &fcxp->hcb_qe, __bfa_fcxp_send_cbfn, fcxp); } } } static struct bfa_fcxp_s * bfa_fcxp_get(struct bfa_fcxp_mod_s *fm, bfa_boolean_t req) { struct bfa_fcxp_s *fcxp; if (req) bfa_q_deq(&fm->fcxp_req_free_q, &fcxp); else bfa_q_deq(&fm->fcxp_rsp_free_q, &fcxp); if (fcxp) list_add_tail(&fcxp->qe, &fm->fcxp_active_q); return fcxp; } static void bfa_fcxp_init_reqrsp(struct bfa_fcxp_s *fcxp, struct bfa_s *bfa, u8 *use_ibuf, u32 *nr_sgles, bfa_fcxp_get_sgaddr_t *r_sga_cbfn, bfa_fcxp_get_sglen_t *r_sglen_cbfn, struct list_head *r_sgpg_q, int n_sgles, bfa_fcxp_get_sgaddr_t sga_cbfn, bfa_fcxp_get_sglen_t sglen_cbfn) { WARN_ON(bfa == NULL); bfa_trc(bfa, fcxp->fcxp_tag); if (n_sgles == 0) { *use_ibuf = 1; } else { WARN_ON(*sga_cbfn == NULL); WARN_ON(*sglen_cbfn == NULL); *use_ibuf = 0; *r_sga_cbfn = sga_cbfn; *r_sglen_cbfn = sglen_cbfn; *nr_sgles = n_sgles; /* * alloc required sgpgs */ if (n_sgles > BFI_SGE_INLINE) WARN_ON(1); } } static void bfa_fcxp_init(struct bfa_fcxp_s *fcxp, void *caller, struct bfa_s *bfa, int nreq_sgles, int nrsp_sgles, bfa_fcxp_get_sgaddr_t req_sga_cbfn, bfa_fcxp_get_sglen_t req_sglen_cbfn, bfa_fcxp_get_sgaddr_t rsp_sga_cbfn, bfa_fcxp_get_sglen_t rsp_sglen_cbfn) { WARN_ON(bfa == NULL); bfa_trc(bfa, fcxp->fcxp_tag); fcxp->caller = caller; bfa_fcxp_init_reqrsp(fcxp, bfa, &fcxp->use_ireqbuf, &fcxp->nreq_sgles, &fcxp->req_sga_cbfn, &fcxp->req_sglen_cbfn, &fcxp->req_sgpg_q, nreq_sgles, req_sga_cbfn, req_sglen_cbfn); bfa_fcxp_init_reqrsp(fcxp, bfa, &fcxp->use_irspbuf, &fcxp->nrsp_sgles, &fcxp->rsp_sga_cbfn, &fcxp->rsp_sglen_cbfn, &fcxp->rsp_sgpg_q, nrsp_sgles, rsp_sga_cbfn, rsp_sglen_cbfn); } static void bfa_fcxp_put(struct bfa_fcxp_s *fcxp) { struct bfa_fcxp_mod_s *mod = fcxp->fcxp_mod; struct bfa_fcxp_wqe_s *wqe; if (fcxp->req_rsp) bfa_q_deq(&mod->req_wait_q, &wqe); else bfa_q_deq(&mod->rsp_wait_q, &wqe); if (wqe) { bfa_trc(mod->bfa, fcxp->fcxp_tag); bfa_fcxp_init(fcxp, wqe->caller, wqe->bfa, wqe->nreq_sgles, wqe->nrsp_sgles, wqe->req_sga_cbfn, wqe->req_sglen_cbfn, wqe->rsp_sga_cbfn, wqe->rsp_sglen_cbfn); wqe->alloc_cbfn(wqe->alloc_cbarg, fcxp); return; } WARN_ON(!bfa_q_is_on_q(&mod->fcxp_active_q, fcxp)); list_del(&fcxp->qe); if (fcxp->req_rsp) list_add_tail(&fcxp->qe, &mod->fcxp_req_free_q); else list_add_tail(&fcxp->qe, &mod->fcxp_rsp_free_q); } static void bfa_fcxp_null_comp(void *bfad_fcxp, struct bfa_fcxp_s *fcxp, void *cbarg, bfa_status_t req_status, u32 rsp_len, u32 resid_len, struct fchs_s *rsp_fchs) { /* discarded fcxp completion */ } static void __bfa_fcxp_send_cbfn(void *cbarg, bfa_boolean_t complete) { struct bfa_fcxp_s *fcxp = cbarg; if (complete) { fcxp->send_cbfn(fcxp->caller, fcxp, fcxp->send_cbarg, fcxp->rsp_status, fcxp->rsp_len, fcxp->residue_len, &fcxp->rsp_fchs); } else { bfa_fcxp_free(fcxp); } } static void hal_fcxp_send_comp(struct bfa_s *bfa, struct bfi_fcxp_send_rsp_s *fcxp_rsp) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); struct bfa_fcxp_s *fcxp; u16 fcxp_tag = be16_to_cpu(fcxp_rsp->fcxp_tag); bfa_trc(bfa, fcxp_tag); fcxp_rsp->rsp_len = be32_to_cpu(fcxp_rsp->rsp_len); /* * @todo f/w should not set residue to non-0 when everything * is received. */ if (fcxp_rsp->req_status == BFA_STATUS_OK) fcxp_rsp->residue_len = 0; else fcxp_rsp->residue_len = be32_to_cpu(fcxp_rsp->residue_len); fcxp = BFA_FCXP_FROM_TAG(mod, fcxp_tag); WARN_ON(fcxp->send_cbfn == NULL); hal_fcxp_rx_plog(mod->bfa, fcxp, fcxp_rsp); if (fcxp->send_cbfn != NULL) { bfa_trc(mod->bfa, (NULL == fcxp->caller)); if (fcxp->caller == NULL) { fcxp->send_cbfn(fcxp->caller, fcxp, fcxp->send_cbarg, fcxp_rsp->req_status, fcxp_rsp->rsp_len, fcxp_rsp->residue_len, &fcxp_rsp->fchs); /* * fcxp automatically freed on return from the callback */ bfa_fcxp_free(fcxp); } else { fcxp->rsp_status = fcxp_rsp->req_status; fcxp->rsp_len = fcxp_rsp->rsp_len; fcxp->residue_len = fcxp_rsp->residue_len; fcxp->rsp_fchs = fcxp_rsp->fchs; bfa_cb_queue(bfa, &fcxp->hcb_qe, __bfa_fcxp_send_cbfn, fcxp); } } else { bfa_trc(bfa, (NULL == fcxp->send_cbfn)); } } static void hal_fcxp_tx_plog(struct bfa_s *bfa, u32 reqlen, struct bfa_fcxp_s *fcxp, struct fchs_s *fchs) { /* * TODO: TX ox_id */ if (reqlen > 0) { if (fcxp->use_ireqbuf) { u32 pld_w0 = *((u32 *) BFA_FCXP_REQ_PLD(fcxp)); bfa_plog_fchdr_and_pl(bfa->plog, BFA_PL_MID_HAL_FCXP, BFA_PL_EID_TX, reqlen + sizeof(struct fchs_s), fchs, pld_w0); } else { bfa_plog_fchdr(bfa->plog, BFA_PL_MID_HAL_FCXP, BFA_PL_EID_TX, reqlen + sizeof(struct fchs_s), fchs); } } else { bfa_plog_fchdr(bfa->plog, BFA_PL_MID_HAL_FCXP, BFA_PL_EID_TX, reqlen + sizeof(struct fchs_s), fchs); } } static void hal_fcxp_rx_plog(struct bfa_s *bfa, struct bfa_fcxp_s *fcxp, struct bfi_fcxp_send_rsp_s *fcxp_rsp) { if (fcxp_rsp->rsp_len > 0) { if (fcxp->use_irspbuf) { u32 pld_w0 = *((u32 *) BFA_FCXP_RSP_PLD(fcxp)); bfa_plog_fchdr_and_pl(bfa->plog, BFA_PL_MID_HAL_FCXP, BFA_PL_EID_RX, (u16) fcxp_rsp->rsp_len, &fcxp_rsp->fchs, pld_w0); } else { bfa_plog_fchdr(bfa->plog, BFA_PL_MID_HAL_FCXP, BFA_PL_EID_RX, (u16) fcxp_rsp->rsp_len, &fcxp_rsp->fchs); } } else { bfa_plog_fchdr(bfa->plog, BFA_PL_MID_HAL_FCXP, BFA_PL_EID_RX, (u16) fcxp_rsp->rsp_len, &fcxp_rsp->fchs); } } /* * Handler to resume sending fcxp when space in available in cpe queue. */ static void bfa_fcxp_qresume(void *cbarg) { struct bfa_fcxp_s *fcxp = cbarg; struct bfa_s *bfa = fcxp->fcxp_mod->bfa; struct bfi_fcxp_send_req_s *send_req; fcxp->reqq_waiting = BFA_FALSE; send_req = bfa_reqq_next(bfa, BFA_REQQ_FCXP); bfa_fcxp_queue(fcxp, send_req); } /* * Queue fcxp send request to foimrware. */ static void bfa_fcxp_queue(struct bfa_fcxp_s *fcxp, struct bfi_fcxp_send_req_s *send_req) { struct bfa_s *bfa = fcxp->fcxp_mod->bfa; struct bfa_fcxp_req_info_s *reqi = &fcxp->req_info; struct bfa_fcxp_rsp_info_s *rspi = &fcxp->rsp_info; struct bfa_rport_s *rport = reqi->bfa_rport; bfi_h2i_set(send_req->mh, BFI_MC_FCXP, BFI_FCXP_H2I_SEND_REQ, bfa_fn_lpu(bfa)); send_req->fcxp_tag = cpu_to_be16(fcxp->fcxp_tag); if (rport) { send_req->rport_fw_hndl = rport->fw_handle; send_req->max_frmsz = cpu_to_be16(rport->rport_info.max_frmsz); if (send_req->max_frmsz == 0) send_req->max_frmsz = cpu_to_be16(FC_MAX_PDUSZ); } else { send_req->rport_fw_hndl = 0; send_req->max_frmsz = cpu_to_be16(FC_MAX_PDUSZ); } send_req->vf_id = cpu_to_be16(reqi->vf_id); send_req->lp_fwtag = bfa_lps_get_fwtag(bfa, reqi->lp_tag); send_req->class = reqi->class; send_req->rsp_timeout = rspi->rsp_timeout; send_req->cts = reqi->cts; send_req->fchs = reqi->fchs; send_req->req_len = cpu_to_be32(reqi->req_tot_len); send_req->rsp_maxlen = cpu_to_be32(rspi->rsp_maxlen); /* * setup req sgles */ if (fcxp->use_ireqbuf == 1) { bfa_alen_set(&send_req->req_alen, reqi->req_tot_len, BFA_FCXP_REQ_PLD_PA(fcxp)); } else { if (fcxp->nreq_sgles > 0) { WARN_ON(fcxp->nreq_sgles != 1); bfa_alen_set(&send_req->req_alen, reqi->req_tot_len, fcxp->req_sga_cbfn(fcxp->caller, 0)); } else { WARN_ON(reqi->req_tot_len != 0); bfa_alen_set(&send_req->rsp_alen, 0, 0); } } /* * setup rsp sgles */ if (fcxp->use_irspbuf == 1) { WARN_ON(rspi->rsp_maxlen > BFA_FCXP_MAX_LBUF_SZ); bfa_alen_set(&send_req->rsp_alen, rspi->rsp_maxlen, BFA_FCXP_RSP_PLD_PA(fcxp)); } else { if (fcxp->nrsp_sgles > 0) { WARN_ON(fcxp->nrsp_sgles != 1); bfa_alen_set(&send_req->rsp_alen, rspi->rsp_maxlen, fcxp->rsp_sga_cbfn(fcxp->caller, 0)); } else { WARN_ON(rspi->rsp_maxlen != 0); bfa_alen_set(&send_req->rsp_alen, 0, 0); } } hal_fcxp_tx_plog(bfa, reqi->req_tot_len, fcxp, &reqi->fchs); bfa_reqq_produce(bfa, BFA_REQQ_FCXP, send_req->mh); bfa_trc(bfa, bfa_reqq_pi(bfa, BFA_REQQ_FCXP)); bfa_trc(bfa, bfa_reqq_ci(bfa, BFA_REQQ_FCXP)); } /* * Allocate an FCXP instance to send a response or to send a request * that has a response. Request/response buffers are allocated by caller. * * @param[in] bfa BFA bfa instance * @param[in] nreq_sgles Number of SG elements required for request * buffer. 0, if fcxp internal buffers are used. * Use bfa_fcxp_get_reqbuf() to get the * internal req buffer. * @param[in] req_sgles SG elements describing request buffer. Will be * copied in by BFA and hence can be freed on * return from this function. * @param[in] get_req_sga function ptr to be called to get a request SG * Address (given the sge index). * @param[in] get_req_sglen function ptr to be called to get a request SG * len (given the sge index). * @param[in] get_rsp_sga function ptr to be called to get a response SG * Address (given the sge index). * @param[in] get_rsp_sglen function ptr to be called to get a response SG * len (given the sge index). * @param[in] req Allocated FCXP is used to send req or rsp? * request - BFA_TRUE, response - BFA_FALSE * * @return FCXP instance. NULL on failure. */ struct bfa_fcxp_s * bfa_fcxp_req_rsp_alloc(void *caller, struct bfa_s *bfa, int nreq_sgles, int nrsp_sgles, bfa_fcxp_get_sgaddr_t req_sga_cbfn, bfa_fcxp_get_sglen_t req_sglen_cbfn, bfa_fcxp_get_sgaddr_t rsp_sga_cbfn, bfa_fcxp_get_sglen_t rsp_sglen_cbfn, bfa_boolean_t req) { struct bfa_fcxp_s *fcxp = NULL; WARN_ON(bfa == NULL); fcxp = bfa_fcxp_get(BFA_FCXP_MOD(bfa), req); if (fcxp == NULL) return NULL; bfa_trc(bfa, fcxp->fcxp_tag); bfa_fcxp_init(fcxp, caller, bfa, nreq_sgles, nrsp_sgles, req_sga_cbfn, req_sglen_cbfn, rsp_sga_cbfn, rsp_sglen_cbfn); return fcxp; } /* * Get the internal request buffer pointer * * @param[in] fcxp BFA fcxp pointer * * @return pointer to the internal request buffer */ void * bfa_fcxp_get_reqbuf(struct bfa_fcxp_s *fcxp) { struct bfa_fcxp_mod_s *mod = fcxp->fcxp_mod; void *reqbuf; WARN_ON(fcxp->use_ireqbuf != 1); reqbuf = bfa_mem_get_dmabuf_kva(mod, fcxp->fcxp_tag, mod->req_pld_sz + mod->rsp_pld_sz); return reqbuf; } u32 bfa_fcxp_get_reqbufsz(struct bfa_fcxp_s *fcxp) { struct bfa_fcxp_mod_s *mod = fcxp->fcxp_mod; return mod->req_pld_sz; } /* * Get the internal response buffer pointer * * @param[in] fcxp BFA fcxp pointer * * @return pointer to the internal request buffer */ void * bfa_fcxp_get_rspbuf(struct bfa_fcxp_s *fcxp) { struct bfa_fcxp_mod_s *mod = fcxp->fcxp_mod; void *fcxp_buf; WARN_ON(fcxp->use_irspbuf != 1); fcxp_buf = bfa_mem_get_dmabuf_kva(mod, fcxp->fcxp_tag, mod->req_pld_sz + mod->rsp_pld_sz); /* fcxp_buf = req_buf + rsp_buf :- add req_buf_sz to get to rsp_buf */ return ((u8 *) fcxp_buf) + mod->req_pld_sz; } /* * Free the BFA FCXP * * @param[in] fcxp BFA fcxp pointer * * @return void */ void bfa_fcxp_free(struct bfa_fcxp_s *fcxp) { struct bfa_fcxp_mod_s *mod = fcxp->fcxp_mod; WARN_ON(fcxp == NULL); bfa_trc(mod->bfa, fcxp->fcxp_tag); bfa_fcxp_put(fcxp); } /* * Send a FCXP request * * @param[in] fcxp BFA fcxp pointer * @param[in] rport BFA rport pointer. Could be left NULL for WKA rports * @param[in] vf_id virtual Fabric ID * @param[in] lp_tag lport tag * @param[in] cts use Continuous sequence * @param[in] cos fc Class of Service * @param[in] reqlen request length, does not include FCHS length * @param[in] fchs fc Header Pointer. The header content will be copied * in by BFA. * * @param[in] cbfn call back function to be called on receiving * the response * @param[in] cbarg arg for cbfn * @param[in] rsp_timeout * response timeout * * @return bfa_status_t */ void bfa_fcxp_send(struct bfa_fcxp_s *fcxp, struct bfa_rport_s *rport, u16 vf_id, u8 lp_tag, bfa_boolean_t cts, enum fc_cos cos, u32 reqlen, struct fchs_s *fchs, bfa_cb_fcxp_send_t cbfn, void *cbarg, u32 rsp_maxlen, u8 rsp_timeout) { struct bfa_s *bfa = fcxp->fcxp_mod->bfa; struct bfa_fcxp_req_info_s *reqi = &fcxp->req_info; struct bfa_fcxp_rsp_info_s *rspi = &fcxp->rsp_info; struct bfi_fcxp_send_req_s *send_req; bfa_trc(bfa, fcxp->fcxp_tag); /* * setup request/response info */ reqi->bfa_rport = rport; reqi->vf_id = vf_id; reqi->lp_tag = lp_tag; reqi->class = cos; rspi->rsp_timeout = rsp_timeout; reqi->cts = cts; reqi->fchs = *fchs; reqi->req_tot_len = reqlen; rspi->rsp_maxlen = rsp_maxlen; fcxp->send_cbfn = cbfn ? cbfn : bfa_fcxp_null_comp; fcxp->send_cbarg = cbarg; /* * If no room in CPE queue, wait for space in request queue */ send_req = bfa_reqq_next(bfa, BFA_REQQ_FCXP); if (!send_req) { bfa_trc(bfa, fcxp->fcxp_tag); fcxp->reqq_waiting = BFA_TRUE; bfa_reqq_wait(bfa, BFA_REQQ_FCXP, &fcxp->reqq_wqe); return; } bfa_fcxp_queue(fcxp, send_req); } /* * Abort a BFA FCXP * * @param[in] fcxp BFA fcxp pointer * * @return void */ bfa_status_t bfa_fcxp_abort(struct bfa_fcxp_s *fcxp) { bfa_trc(fcxp->fcxp_mod->bfa, fcxp->fcxp_tag); WARN_ON(1); return BFA_STATUS_OK; } void bfa_fcxp_req_rsp_alloc_wait(struct bfa_s *bfa, struct bfa_fcxp_wqe_s *wqe, bfa_fcxp_alloc_cbfn_t alloc_cbfn, void *alloc_cbarg, void *caller, int nreq_sgles, int nrsp_sgles, bfa_fcxp_get_sgaddr_t req_sga_cbfn, bfa_fcxp_get_sglen_t req_sglen_cbfn, bfa_fcxp_get_sgaddr_t rsp_sga_cbfn, bfa_fcxp_get_sglen_t rsp_sglen_cbfn, bfa_boolean_t req) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); if (req) WARN_ON(!list_empty(&mod->fcxp_req_free_q)); else WARN_ON(!list_empty(&mod->fcxp_rsp_free_q)); wqe->alloc_cbfn = alloc_cbfn; wqe->alloc_cbarg = alloc_cbarg; wqe->caller = caller; wqe->bfa = bfa; wqe->nreq_sgles = nreq_sgles; wqe->nrsp_sgles = nrsp_sgles; wqe->req_sga_cbfn = req_sga_cbfn; wqe->req_sglen_cbfn = req_sglen_cbfn; wqe->rsp_sga_cbfn = rsp_sga_cbfn; wqe->rsp_sglen_cbfn = rsp_sglen_cbfn; if (req) list_add_tail(&wqe->qe, &mod->req_wait_q); else list_add_tail(&wqe->qe, &mod->rsp_wait_q); } void bfa_fcxp_walloc_cancel(struct bfa_s *bfa, struct bfa_fcxp_wqe_s *wqe) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); WARN_ON(!bfa_q_is_on_q(&mod->req_wait_q, wqe) || !bfa_q_is_on_q(&mod->rsp_wait_q, wqe)); list_del(&wqe->qe); } void bfa_fcxp_discard(struct bfa_fcxp_s *fcxp) { /* * If waiting for room in request queue, cancel reqq wait * and free fcxp. */ if (fcxp->reqq_waiting) { fcxp->reqq_waiting = BFA_FALSE; bfa_reqq_wcancel(&fcxp->reqq_wqe); bfa_fcxp_free(fcxp); return; } fcxp->send_cbfn = bfa_fcxp_null_comp; } void bfa_fcxp_isr(struct bfa_s *bfa, struct bfi_msg_s *msg) { switch (msg->mhdr.msg_id) { case BFI_FCXP_I2H_SEND_RSP: hal_fcxp_send_comp(bfa, (struct bfi_fcxp_send_rsp_s *) msg); break; default: bfa_trc(bfa, msg->mhdr.msg_id); WARN_ON(1); } } u32 bfa_fcxp_get_maxrsp(struct bfa_s *bfa) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); return mod->rsp_pld_sz; } void bfa_fcxp_res_recfg(struct bfa_s *bfa, u16 num_fcxp_fw) { struct bfa_fcxp_mod_s *mod = BFA_FCXP_MOD(bfa); struct list_head *qe; int i; for (i = 0; i < (mod->num_fcxps - num_fcxp_fw); i++) { if (i < ((mod->num_fcxps - num_fcxp_fw) / 2)) { bfa_q_deq_tail(&mod->fcxp_req_free_q, &qe); list_add_tail(qe, &mod->fcxp_req_unused_q); } else { bfa_q_deq_tail(&mod->fcxp_rsp_free_q, &qe); list_add_tail(qe, &mod->fcxp_rsp_unused_q); } } } /* * BFA LPS state machine functions */ /* * Init state -- no login */ static void bfa_lps_sm_init(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_LOGIN: if (bfa_reqq_full(lps->bfa, lps->reqq)) { bfa_sm_set_state(lps, bfa_lps_sm_loginwait); bfa_reqq_wait(lps->bfa, lps->reqq, &lps->wqe); } else { bfa_sm_set_state(lps, bfa_lps_sm_login); bfa_lps_send_login(lps); } if (lps->fdisc) bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGIN, 0, "FDISC Request"); else bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGIN, 0, "FLOGI Request"); break; case BFA_LPS_SM_LOGOUT: bfa_lps_logout_comp(lps); break; case BFA_LPS_SM_DELETE: bfa_lps_free(lps); break; case BFA_LPS_SM_RX_CVL: case BFA_LPS_SM_OFFLINE: break; case BFA_LPS_SM_FWRSP: /* * Could happen when fabric detects loopback and discards * the lps request. Fw will eventually sent out the timeout * Just ignore */ break; case BFA_LPS_SM_SET_N2N_PID: /* * When topology is set to loop, bfa_lps_set_n2n_pid() sends * this event. Ignore this event. */ break; default: bfa_sm_fault(lps->bfa, event); } } /* * login is in progress -- awaiting response from firmware */ static void bfa_lps_sm_login(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_FWRSP: if (lps->status == BFA_STATUS_OK) { bfa_sm_set_state(lps, bfa_lps_sm_online); if (lps->fdisc) bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGIN, 0, "FDISC Accept"); else bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGIN, 0, "FLOGI Accept"); /* If N2N, send the assigned PID to FW */ bfa_trc(lps->bfa, lps->fport); bfa_trc(lps->bfa, lps->lp_pid); if (!lps->fport && lps->lp_pid) bfa_sm_send_event(lps, BFA_LPS_SM_SET_N2N_PID); } else { bfa_sm_set_state(lps, bfa_lps_sm_init); if (lps->fdisc) bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGIN, 0, "FDISC Fail (RJT or timeout)"); else bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGIN, 0, "FLOGI Fail (RJT or timeout)"); } bfa_lps_login_comp(lps); break; case BFA_LPS_SM_OFFLINE: case BFA_LPS_SM_DELETE: bfa_sm_set_state(lps, bfa_lps_sm_init); break; case BFA_LPS_SM_SET_N2N_PID: bfa_trc(lps->bfa, lps->fport); bfa_trc(lps->bfa, lps->lp_pid); break; default: bfa_sm_fault(lps->bfa, event); } } /* * login pending - awaiting space in request queue */ static void bfa_lps_sm_loginwait(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_RESUME: bfa_sm_set_state(lps, bfa_lps_sm_login); bfa_lps_send_login(lps); break; case BFA_LPS_SM_OFFLINE: case BFA_LPS_SM_DELETE: bfa_sm_set_state(lps, bfa_lps_sm_init); bfa_reqq_wcancel(&lps->wqe); break; case BFA_LPS_SM_RX_CVL: /* * Login was not even sent out; so when getting out * of this state, it will appear like a login retry * after Clear virtual link */ break; default: bfa_sm_fault(lps->bfa, event); } } /* * login complete */ static void bfa_lps_sm_online(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_LOGOUT: if (bfa_reqq_full(lps->bfa, lps->reqq)) { bfa_sm_set_state(lps, bfa_lps_sm_logowait); bfa_reqq_wait(lps->bfa, lps->reqq, &lps->wqe); } else { bfa_sm_set_state(lps, bfa_lps_sm_logout); bfa_lps_send_logout(lps); } bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGO, 0, "Logout"); break; case BFA_LPS_SM_RX_CVL: bfa_sm_set_state(lps, bfa_lps_sm_init); /* Let the vport module know about this event */ bfa_lps_cvl_event(lps); bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_FIP_FCF_CVL, 0, "FCF Clear Virt. Link Rx"); break; case BFA_LPS_SM_SET_N2N_PID: if (bfa_reqq_full(lps->bfa, lps->reqq)) { bfa_sm_set_state(lps, bfa_lps_sm_online_n2n_pid_wait); bfa_reqq_wait(lps->bfa, lps->reqq, &lps->wqe); } else bfa_lps_send_set_n2n_pid(lps); break; case BFA_LPS_SM_OFFLINE: case BFA_LPS_SM_DELETE: bfa_sm_set_state(lps, bfa_lps_sm_init); break; default: bfa_sm_fault(lps->bfa, event); } } /* * login complete */ static void bfa_lps_sm_online_n2n_pid_wait(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_RESUME: bfa_sm_set_state(lps, bfa_lps_sm_online); bfa_lps_send_set_n2n_pid(lps); break; case BFA_LPS_SM_LOGOUT: bfa_sm_set_state(lps, bfa_lps_sm_logowait); bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_LOGO, 0, "Logout"); break; case BFA_LPS_SM_RX_CVL: bfa_sm_set_state(lps, bfa_lps_sm_init); bfa_reqq_wcancel(&lps->wqe); /* Let the vport module know about this event */ bfa_lps_cvl_event(lps); bfa_plog_str(lps->bfa->plog, BFA_PL_MID_LPS, BFA_PL_EID_FIP_FCF_CVL, 0, "FCF Clear Virt. Link Rx"); break; case BFA_LPS_SM_OFFLINE: case BFA_LPS_SM_DELETE: bfa_sm_set_state(lps, bfa_lps_sm_init); bfa_reqq_wcancel(&lps->wqe); break; default: bfa_sm_fault(lps->bfa, event); } } /* * logout in progress - awaiting firmware response */ static void bfa_lps_sm_logout(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_FWRSP: case BFA_LPS_SM_OFFLINE: bfa_sm_set_state(lps, bfa_lps_sm_init); bfa_lps_logout_comp(lps); break; case BFA_LPS_SM_DELETE: bfa_sm_set_state(lps, bfa_lps_sm_init); break; default: bfa_sm_fault(lps->bfa, event); } } /* * logout pending -- awaiting space in request queue */ static void bfa_lps_sm_logowait(struct bfa_lps_s *lps, enum bfa_lps_event event) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, event); switch (event) { case BFA_LPS_SM_RESUME: bfa_sm_set_state(lps, bfa_lps_sm_logout); bfa_lps_send_logout(lps); break; case BFA_LPS_SM_OFFLINE: case BFA_LPS_SM_DELETE: bfa_sm_set_state(lps, bfa_lps_sm_init); bfa_reqq_wcancel(&lps->wqe); break; default: bfa_sm_fault(lps->bfa, event); } } /* * lps_pvt BFA LPS private functions */ /* * return memory requirement */ void bfa_lps_meminfo(struct bfa_iocfc_cfg_s *cfg, struct bfa_meminfo_s *minfo, struct bfa_s *bfa) { struct bfa_mem_kva_s *lps_kva = BFA_MEM_LPS_KVA(bfa); if (cfg->drvcfg.min_cfg) bfa_mem_kva_setup(minfo, lps_kva, sizeof(struct bfa_lps_s) * BFA_LPS_MIN_LPORTS); else bfa_mem_kva_setup(minfo, lps_kva, sizeof(struct bfa_lps_s) * BFA_LPS_MAX_LPORTS); } /* * bfa module attach at initialization time */ void bfa_lps_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps; int i; mod->num_lps = BFA_LPS_MAX_LPORTS; if (cfg->drvcfg.min_cfg) mod->num_lps = BFA_LPS_MIN_LPORTS; else mod->num_lps = BFA_LPS_MAX_LPORTS; mod->lps_arr = lps = (struct bfa_lps_s *) bfa_mem_kva_curp(mod); bfa_mem_kva_curp(mod) += mod->num_lps * sizeof(struct bfa_lps_s); INIT_LIST_HEAD(&mod->lps_free_q); INIT_LIST_HEAD(&mod->lps_active_q); INIT_LIST_HEAD(&mod->lps_login_q); for (i = 0; i < mod->num_lps; i++, lps++) { lps->bfa = bfa; lps->bfa_tag = (u8) i; lps->reqq = BFA_REQQ_LPS; bfa_reqq_winit(&lps->wqe, bfa_lps_reqq_resume, lps); list_add_tail(&lps->qe, &mod->lps_free_q); } } /* * IOC in disabled state -- consider all lps offline */ void bfa_lps_iocdisable(struct bfa_s *bfa) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps; struct list_head *qe, *qen; list_for_each_safe(qe, qen, &mod->lps_active_q) { lps = (struct bfa_lps_s *) qe; bfa_sm_send_event(lps, BFA_LPS_SM_OFFLINE); } list_for_each_safe(qe, qen, &mod->lps_login_q) { lps = (struct bfa_lps_s *) qe; bfa_sm_send_event(lps, BFA_LPS_SM_OFFLINE); } list_splice_tail_init(&mod->lps_login_q, &mod->lps_active_q); } /* * Firmware login response */ static void bfa_lps_login_rsp(struct bfa_s *bfa, struct bfi_lps_login_rsp_s *rsp) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps; WARN_ON(rsp->bfa_tag >= mod->num_lps); lps = BFA_LPS_FROM_TAG(mod, rsp->bfa_tag); lps->status = rsp->status; switch (rsp->status) { case BFA_STATUS_OK: lps->fw_tag = rsp->fw_tag; lps->fport = rsp->f_port; if (lps->fport) lps->lp_pid = rsp->lp_pid; lps->npiv_en = rsp->npiv_en; lps->pr_bbcred = be16_to_cpu(rsp->bb_credit); lps->pr_pwwn = rsp->port_name; lps->pr_nwwn = rsp->node_name; lps->auth_req = rsp->auth_req; lps->lp_mac = rsp->lp_mac; lps->brcd_switch = rsp->brcd_switch; lps->fcf_mac = rsp->fcf_mac; break; case BFA_STATUS_FABRIC_RJT: lps->lsrjt_rsn = rsp->lsrjt_rsn; lps->lsrjt_expl = rsp->lsrjt_expl; break; case BFA_STATUS_EPROTOCOL: lps->ext_status = rsp->ext_status; break; case BFA_STATUS_VPORT_MAX: if (rsp->ext_status) bfa_lps_no_res(lps, rsp->ext_status); break; default: /* Nothing to do with other status */ break; } list_del(&lps->qe); list_add_tail(&lps->qe, &mod->lps_active_q); bfa_sm_send_event(lps, BFA_LPS_SM_FWRSP); } static void bfa_lps_no_res(struct bfa_lps_s *first_lps, u8 count) { struct bfa_s *bfa = first_lps->bfa; struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct list_head *qe, *qe_next; struct bfa_lps_s *lps; bfa_trc(bfa, count); qe = bfa_q_next(first_lps); while (count && qe) { qe_next = bfa_q_next(qe); lps = (struct bfa_lps_s *)qe; bfa_trc(bfa, lps->bfa_tag); lps->status = first_lps->status; list_del(&lps->qe); list_add_tail(&lps->qe, &mod->lps_active_q); bfa_sm_send_event(lps, BFA_LPS_SM_FWRSP); qe = qe_next; count--; } } /* * Firmware logout response */ static void bfa_lps_logout_rsp(struct bfa_s *bfa, struct bfi_lps_logout_rsp_s *rsp) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps; WARN_ON(rsp->bfa_tag >= mod->num_lps); lps = BFA_LPS_FROM_TAG(mod, rsp->bfa_tag); bfa_sm_send_event(lps, BFA_LPS_SM_FWRSP); } /* * Firmware received a Clear virtual link request (for FCoE) */ static void bfa_lps_rx_cvl_event(struct bfa_s *bfa, struct bfi_lps_cvl_event_s *cvl) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps; lps = BFA_LPS_FROM_TAG(mod, cvl->bfa_tag); bfa_sm_send_event(lps, BFA_LPS_SM_RX_CVL); } /* * Space is available in request queue, resume queueing request to firmware. */ static void bfa_lps_reqq_resume(void *lps_arg) { struct bfa_lps_s *lps = lps_arg; bfa_sm_send_event(lps, BFA_LPS_SM_RESUME); } /* * lps is freed -- triggered by vport delete */ static void bfa_lps_free(struct bfa_lps_s *lps) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(lps->bfa); lps->lp_pid = 0; list_del(&lps->qe); list_add_tail(&lps->qe, &mod->lps_free_q); } /* * send login request to firmware */ static void bfa_lps_send_login(struct bfa_lps_s *lps) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(lps->bfa); struct bfi_lps_login_req_s *m; m = bfa_reqq_next(lps->bfa, lps->reqq); WARN_ON(!m); bfi_h2i_set(m->mh, BFI_MC_LPS, BFI_LPS_H2I_LOGIN_REQ, bfa_fn_lpu(lps->bfa)); m->bfa_tag = lps->bfa_tag; m->alpa = lps->alpa; m->pdu_size = cpu_to_be16(lps->pdusz); m->pwwn = lps->pwwn; m->nwwn = lps->nwwn; m->fdisc = lps->fdisc; m->auth_en = lps->auth_en; bfa_reqq_produce(lps->bfa, lps->reqq, m->mh); list_del(&lps->qe); list_add_tail(&lps->qe, &mod->lps_login_q); } /* * send logout request to firmware */ static void bfa_lps_send_logout(struct bfa_lps_s *lps) { struct bfi_lps_logout_req_s *m; m = bfa_reqq_next(lps->bfa, lps->reqq); WARN_ON(!m); bfi_h2i_set(m->mh, BFI_MC_LPS, BFI_LPS_H2I_LOGOUT_REQ, bfa_fn_lpu(lps->bfa)); m->fw_tag = lps->fw_tag; m->port_name = lps->pwwn; bfa_reqq_produce(lps->bfa, lps->reqq, m->mh); } /* * send n2n pid set request to firmware */ static void bfa_lps_send_set_n2n_pid(struct bfa_lps_s *lps) { struct bfi_lps_n2n_pid_req_s *m; m = bfa_reqq_next(lps->bfa, lps->reqq); WARN_ON(!m); bfi_h2i_set(m->mh, BFI_MC_LPS, BFI_LPS_H2I_N2N_PID_REQ, bfa_fn_lpu(lps->bfa)); m->fw_tag = lps->fw_tag; m->lp_pid = lps->lp_pid; bfa_reqq_produce(lps->bfa, lps->reqq, m->mh); } /* * Indirect login completion handler for non-fcs */ static void bfa_lps_login_comp_cb(void *arg, bfa_boolean_t complete) { struct bfa_lps_s *lps = arg; if (!complete) return; if (lps->fdisc) bfa_cb_lps_fdisc_comp(lps->bfa->bfad, lps->uarg, lps->status); else bfa_cb_lps_flogi_comp(lps->bfa->bfad, lps->uarg, lps->status); } /* * Login completion handler -- direct call for fcs, queue for others */ static void bfa_lps_login_comp(struct bfa_lps_s *lps) { if (!lps->bfa->fcs) { bfa_cb_queue(lps->bfa, &lps->hcb_qe, bfa_lps_login_comp_cb, lps); return; } if (lps->fdisc) bfa_cb_lps_fdisc_comp(lps->bfa->bfad, lps->uarg, lps->status); else bfa_cb_lps_flogi_comp(lps->bfa->bfad, lps->uarg, lps->status); } /* * Indirect logout completion handler for non-fcs */ static void bfa_lps_logout_comp_cb(void *arg, bfa_boolean_t complete) { struct bfa_lps_s *lps = arg; if (!complete) return; if (lps->fdisc) bfa_cb_lps_fdisclogo_comp(lps->bfa->bfad, lps->uarg); else bfa_cb_lps_flogo_comp(lps->bfa->bfad, lps->uarg); } /* * Logout completion handler -- direct call for fcs, queue for others */ static void bfa_lps_logout_comp(struct bfa_lps_s *lps) { if (!lps->bfa->fcs) { bfa_cb_queue(lps->bfa, &lps->hcb_qe, bfa_lps_logout_comp_cb, lps); return; } if (lps->fdisc) bfa_cb_lps_fdisclogo_comp(lps->bfa->bfad, lps->uarg); } /* * Clear virtual link completion handler for non-fcs */ static void bfa_lps_cvl_event_cb(void *arg, bfa_boolean_t complete) { struct bfa_lps_s *lps = arg; if (!complete) return; /* Clear virtual link to base port will result in link down */ if (lps->fdisc) bfa_cb_lps_cvl_event(lps->bfa->bfad, lps->uarg); } /* * Received Clear virtual link event --direct call for fcs, * queue for others */ static void bfa_lps_cvl_event(struct bfa_lps_s *lps) { if (!lps->bfa->fcs) { bfa_cb_queue(lps->bfa, &lps->hcb_qe, bfa_lps_cvl_event_cb, lps); return; } /* Clear virtual link to base port will result in link down */ if (lps->fdisc) bfa_cb_lps_cvl_event(lps->bfa->bfad, lps->uarg); } /* * lps_public BFA LPS public functions */ u32 bfa_lps_get_max_vport(struct bfa_s *bfa) { if (bfa_ioc_devid(&bfa->ioc) == BFA_PCI_DEVICE_ID_CT) return BFA_LPS_MAX_VPORTS_SUPP_CT; else return BFA_LPS_MAX_VPORTS_SUPP_CB; } /* * Allocate a lport srvice tag. */ struct bfa_lps_s * bfa_lps_alloc(struct bfa_s *bfa) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps = NULL; bfa_q_deq(&mod->lps_free_q, &lps); if (lps == NULL) return NULL; list_add_tail(&lps->qe, &mod->lps_active_q); bfa_sm_set_state(lps, bfa_lps_sm_init); return lps; } /* * Free lport service tag. This can be called anytime after an alloc. * No need to wait for any pending login/logout completions. */ void bfa_lps_delete(struct bfa_lps_s *lps) { bfa_sm_send_event(lps, BFA_LPS_SM_DELETE); } /* * Initiate a lport login. */ void bfa_lps_flogi(struct bfa_lps_s *lps, void *uarg, u8 alpa, u16 pdusz, wwn_t pwwn, wwn_t nwwn, bfa_boolean_t auth_en) { lps->uarg = uarg; lps->alpa = alpa; lps->pdusz = pdusz; lps->pwwn = pwwn; lps->nwwn = nwwn; lps->fdisc = BFA_FALSE; lps->auth_en = auth_en; bfa_sm_send_event(lps, BFA_LPS_SM_LOGIN); } /* * Initiate a lport fdisc login. */ void bfa_lps_fdisc(struct bfa_lps_s *lps, void *uarg, u16 pdusz, wwn_t pwwn, wwn_t nwwn) { lps->uarg = uarg; lps->alpa = 0; lps->pdusz = pdusz; lps->pwwn = pwwn; lps->nwwn = nwwn; lps->fdisc = BFA_TRUE; lps->auth_en = BFA_FALSE; bfa_sm_send_event(lps, BFA_LPS_SM_LOGIN); } /* * Initiate a lport FDSIC logout. */ void bfa_lps_fdisclogo(struct bfa_lps_s *lps) { bfa_sm_send_event(lps, BFA_LPS_SM_LOGOUT); } u8 bfa_lps_get_fwtag(struct bfa_s *bfa, u8 lp_tag) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); return BFA_LPS_FROM_TAG(mod, lp_tag)->fw_tag; } /* * Return lport services tag given the pid */ u8 bfa_lps_get_tag_from_pid(struct bfa_s *bfa, u32 pid) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); struct bfa_lps_s *lps; int i; for (i = 0, lps = mod->lps_arr; i < mod->num_lps; i++, lps++) { if (lps->lp_pid == pid) return lps->bfa_tag; } /* Return base port tag anyway */ return 0; } /* * return port id assigned to the base lport */ u32 bfa_lps_get_base_pid(struct bfa_s *bfa) { struct bfa_lps_mod_s *mod = BFA_LPS_MOD(bfa); return BFA_LPS_FROM_TAG(mod, 0)->lp_pid; } /* * Set PID in case of n2n (which is assigned during PLOGI) */ void bfa_lps_set_n2n_pid(struct bfa_lps_s *lps, uint32_t n2n_pid) { bfa_trc(lps->bfa, lps->bfa_tag); bfa_trc(lps->bfa, n2n_pid); lps->lp_pid = n2n_pid; bfa_sm_send_event(lps, BFA_LPS_SM_SET_N2N_PID); } /* * LPS firmware message class handler. */ void bfa_lps_isr(struct bfa_s *bfa, struct bfi_msg_s *m) { union bfi_lps_i2h_msg_u msg; bfa_trc(bfa, m->mhdr.msg_id); msg.msg = m; switch (m->mhdr.msg_id) { case BFI_LPS_I2H_LOGIN_RSP: bfa_lps_login_rsp(bfa, msg.login_rsp); break; case BFI_LPS_I2H_LOGOUT_RSP: bfa_lps_logout_rsp(bfa, msg.logout_rsp); break; case BFI_LPS_I2H_CVL_EVENT: bfa_lps_rx_cvl_event(bfa, msg.cvl_event); break; default: bfa_trc(bfa, m->mhdr.msg_id); WARN_ON(1); } } static void bfa_fcport_aen_post(struct bfa_fcport_s *fcport, enum bfa_port_aen_event event) { struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; struct bfa_aen_entry_s *aen_entry; bfad_get_aen_entry(bfad, aen_entry); if (!aen_entry) return; aen_entry->aen_data.port.ioc_type = bfa_get_type(fcport->bfa); aen_entry->aen_data.port.pwwn = fcport->pwwn; /* Send the AEN notification */ bfad_im_post_vendor_event(aen_entry, bfad, ++fcport->bfa->bfa_aen_seq, BFA_AEN_CAT_PORT, event); } /* * FC PORT state machine functions */ static void bfa_fcport_sm_uninit(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_START: /* * Start event after IOC is configured and BFA is started. */ fcport->use_flash_cfg = BFA_TRUE; if (bfa_fcport_send_enable(fcport)) { bfa_trc(fcport->bfa, BFA_TRUE); bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); } else { bfa_trc(fcport->bfa, BFA_FALSE); bfa_sm_set_state(fcport, bfa_fcport_sm_enabling_qwait); } break; case BFA_FCPORT_SM_ENABLE: /* * Port is persistently configured to be in enabled state. Do * not change state. Port enabling is done when START event is * received. */ break; case BFA_FCPORT_SM_DISABLE: /* * If a port is persistently configured to be disabled, the * first event will a port disable request. */ bfa_sm_set_state(fcport, bfa_fcport_sm_disabled); break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_enabling_qwait(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { char pwwn_buf[BFA_STRING_32]; struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_QRESUME: bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); bfa_fcport_send_enable(fcport); break; case BFA_FCPORT_SM_STOP: bfa_reqq_wcancel(&fcport->reqq_wait); bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_ENABLE: /* * Already enable is in progress. */ break; case BFA_FCPORT_SM_DISABLE: /* * Just send disable request to firmware when room becomes * available in request queue. */ bfa_sm_set_state(fcport, bfa_fcport_sm_disabled); bfa_reqq_wcancel(&fcport->reqq_wait); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_DISABLE, 0, "Port Disable"); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port disabled: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISABLE); break; case BFA_FCPORT_SM_LINKUP: case BFA_FCPORT_SM_LINKDOWN: /* * Possible to get link events when doing back-to-back * enable/disables. */ break; case BFA_FCPORT_SM_HWFAIL: bfa_reqq_wcancel(&fcport->reqq_wait); bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); break; case BFA_FCPORT_SM_FAA_MISCONFIG: bfa_fcport_reset_linkinfo(fcport); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); bfa_sm_set_state(fcport, bfa_fcport_sm_faa_misconfig); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_enabling(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { char pwwn_buf[BFA_STRING_32]; struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_FWRSP: case BFA_FCPORT_SM_LINKDOWN: bfa_sm_set_state(fcport, bfa_fcport_sm_linkdown); break; case BFA_FCPORT_SM_LINKUP: bfa_fcport_update_linkinfo(fcport); bfa_sm_set_state(fcport, bfa_fcport_sm_linkup); WARN_ON(!fcport->event_cbfn); bfa_fcport_scn(fcport, BFA_PORT_LINKUP, BFA_FALSE); break; case BFA_FCPORT_SM_ENABLE: /* * Already being enabled. */ break; case BFA_FCPORT_SM_DISABLE: if (bfa_fcport_send_disable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_disabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_disabling_qwait); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_DISABLE, 0, "Port Disable"); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port disabled: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISABLE); break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); break; case BFA_FCPORT_SM_FAA_MISCONFIG: bfa_fcport_reset_linkinfo(fcport); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); bfa_sm_set_state(fcport, bfa_fcport_sm_faa_misconfig); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_linkdown(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { struct bfi_fcport_event_s *pevent = fcport->event_arg.i2hmsg.event; char pwwn_buf[BFA_STRING_32]; struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_LINKUP: bfa_fcport_update_linkinfo(fcport); bfa_sm_set_state(fcport, bfa_fcport_sm_linkup); WARN_ON(!fcport->event_cbfn); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_ST_CHANGE, 0, "Port Linkup"); if (!bfa_ioc_get_fcmode(&fcport->bfa->ioc)) { bfa_trc(fcport->bfa, pevent->link_state.attr.vc_fcf.fcf.fipenabled); bfa_trc(fcport->bfa, pevent->link_state.attr.vc_fcf.fcf.fipfailed); if (pevent->link_state.attr.vc_fcf.fcf.fipfailed) bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_FIP_FCF_DISC, 0, "FIP FCF Discovery Failed"); else bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_FIP_FCF_DISC, 0, "FIP FCF Discovered"); } bfa_fcport_scn(fcport, BFA_PORT_LINKUP, BFA_FALSE); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port online: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_ONLINE); /* If QoS is enabled and it is not online, send AEN */ if (fcport->cfg.qos_enabled && fcport->qos_attr.state != BFA_QOS_ONLINE) bfa_fcport_aen_post(fcport, BFA_PORT_AEN_QOS_NEG); break; case BFA_FCPORT_SM_LINKDOWN: /* * Possible to get link down event. */ break; case BFA_FCPORT_SM_ENABLE: /* * Already enabled. */ break; case BFA_FCPORT_SM_DISABLE: if (bfa_fcport_send_disable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_disabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_disabling_qwait); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_DISABLE, 0, "Port Disable"); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port disabled: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISABLE); break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); break; case BFA_FCPORT_SM_FAA_MISCONFIG: bfa_fcport_reset_linkinfo(fcport); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); bfa_sm_set_state(fcport, bfa_fcport_sm_faa_misconfig); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_linkup(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { char pwwn_buf[BFA_STRING_32]; struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_ENABLE: /* * Already enabled. */ break; case BFA_FCPORT_SM_DISABLE: if (bfa_fcport_send_disable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_disabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_disabling_qwait); bfa_fcport_reset_linkinfo(fcport); bfa_fcport_scn(fcport, BFA_PORT_LINKDOWN, BFA_FALSE); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_DISABLE, 0, "Port Disable"); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port offline: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_OFFLINE); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port disabled: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISABLE); break; case BFA_FCPORT_SM_LINKDOWN: bfa_sm_set_state(fcport, bfa_fcport_sm_linkdown); bfa_fcport_reset_linkinfo(fcport); bfa_fcport_scn(fcport, BFA_PORT_LINKDOWN, BFA_FALSE); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_ST_CHANGE, 0, "Port Linkdown"); wwn2str(pwwn_buf, fcport->pwwn); if (BFA_PORT_IS_DISABLED(fcport->bfa)) { BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port offline: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_OFFLINE); } else { BFA_LOG(KERN_ERR, bfad, bfa_log_level, "Base port (WWN = %s) " "lost fabric connectivity\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); } break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); bfa_fcport_reset_linkinfo(fcport); wwn2str(pwwn_buf, fcport->pwwn); if (BFA_PORT_IS_DISABLED(fcport->bfa)) { BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port offline: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_OFFLINE); } else { BFA_LOG(KERN_ERR, bfad, bfa_log_level, "Base port (WWN = %s) " "lost fabric connectivity\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); } break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); bfa_fcport_reset_linkinfo(fcport); bfa_fcport_scn(fcport, BFA_PORT_LINKDOWN, BFA_FALSE); wwn2str(pwwn_buf, fcport->pwwn); if (BFA_PORT_IS_DISABLED(fcport->bfa)) { BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port offline: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_OFFLINE); } else { BFA_LOG(KERN_ERR, bfad, bfa_log_level, "Base port (WWN = %s) " "lost fabric connectivity\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); } break; case BFA_FCPORT_SM_FAA_MISCONFIG: bfa_fcport_reset_linkinfo(fcport); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); bfa_sm_set_state(fcport, bfa_fcport_sm_faa_misconfig); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_disabling_qwait(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_QRESUME: bfa_sm_set_state(fcport, bfa_fcport_sm_disabling); bfa_fcport_send_disable(fcport); break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); bfa_reqq_wcancel(&fcport->reqq_wait); break; case BFA_FCPORT_SM_ENABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_toggling_qwait); break; case BFA_FCPORT_SM_DISABLE: /* * Already being disabled. */ break; case BFA_FCPORT_SM_LINKUP: case BFA_FCPORT_SM_LINKDOWN: /* * Possible to get link events when doing back-to-back * enable/disables. */ break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocfail); bfa_reqq_wcancel(&fcport->reqq_wait); break; case BFA_FCPORT_SM_FAA_MISCONFIG: bfa_fcport_reset_linkinfo(fcport); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISCONNECT); bfa_sm_set_state(fcport, bfa_fcport_sm_faa_misconfig); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_toggling_qwait(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_QRESUME: bfa_sm_set_state(fcport, bfa_fcport_sm_disabling); bfa_fcport_send_disable(fcport); if (bfa_fcport_send_enable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_enabling_qwait); break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); bfa_reqq_wcancel(&fcport->reqq_wait); break; case BFA_FCPORT_SM_ENABLE: break; case BFA_FCPORT_SM_DISABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_disabling_qwait); break; case BFA_FCPORT_SM_LINKUP: case BFA_FCPORT_SM_LINKDOWN: /* * Possible to get link events when doing back-to-back * enable/disables. */ break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocfail); bfa_reqq_wcancel(&fcport->reqq_wait); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_disabling(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { char pwwn_buf[BFA_STRING_32]; struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_FWRSP: bfa_sm_set_state(fcport, bfa_fcport_sm_disabled); break; case BFA_FCPORT_SM_DISABLE: /* * Already being disabled. */ break; case BFA_FCPORT_SM_ENABLE: if (bfa_fcport_send_enable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_enabling_qwait); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_ENABLE, 0, "Port Enable"); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port enabled: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_ENABLE); break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_LINKUP: case BFA_FCPORT_SM_LINKDOWN: /* * Possible to get link events when doing back-to-back * enable/disables. */ break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocfail); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_disabled(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { char pwwn_buf[BFA_STRING_32]; struct bfad_s *bfad = (struct bfad_s *)fcport->bfa->bfad; bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_START: /* * Ignore start event for a port that is disabled. */ break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_ENABLE: if (bfa_fcport_send_enable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_enabling_qwait); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_ENABLE, 0, "Port Enable"); wwn2str(pwwn_buf, fcport->pwwn); BFA_LOG(KERN_INFO, bfad, bfa_log_level, "Base port enabled: WWN = %s\n", pwwn_buf); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_ENABLE); break; case BFA_FCPORT_SM_DISABLE: /* * Already disabled. */ break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocfail); break; case BFA_FCPORT_SM_DPORTENABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_dport); break; case BFA_FCPORT_SM_DDPORTENABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_ddport); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_stopped(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_START: if (bfa_fcport_send_enable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_enabling_qwait); break; default: /* * Ignore all other events. */ ; } } /* * Port is enabled. IOC is down/failed. */ static void bfa_fcport_sm_iocdown(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_START: if (bfa_fcport_send_enable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_enabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_enabling_qwait); break; default: /* * Ignore all events. */ ; } } /* * Port is disabled. IOC is down/failed. */ static void bfa_fcport_sm_iocfail(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_START: bfa_sm_set_state(fcport, bfa_fcport_sm_disabled); break; case BFA_FCPORT_SM_ENABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); break; default: /* * Ignore all events. */ ; } } static void bfa_fcport_sm_dport(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_DPORTENABLE: case BFA_FCPORT_SM_DISABLE: case BFA_FCPORT_SM_ENABLE: case BFA_FCPORT_SM_START: /* * Ignore event for a port that is dport */ break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocfail); break; case BFA_FCPORT_SM_DPORTDISABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_disabled); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_ddport(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_DISABLE: case BFA_FCPORT_SM_DDPORTDISABLE: bfa_sm_set_state(fcport, bfa_fcport_sm_disabled); break; case BFA_FCPORT_SM_DPORTENABLE: case BFA_FCPORT_SM_DPORTDISABLE: case BFA_FCPORT_SM_ENABLE: case BFA_FCPORT_SM_START: /** * Ignore event for a port that is ddport */ break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_HWFAIL: bfa_sm_set_state(fcport, bfa_fcport_sm_iocfail); break; default: bfa_sm_fault(fcport->bfa, event); } } static void bfa_fcport_sm_faa_misconfig(struct bfa_fcport_s *fcport, enum bfa_fcport_sm_event event) { bfa_trc(fcport->bfa, event); switch (event) { case BFA_FCPORT_SM_DPORTENABLE: case BFA_FCPORT_SM_ENABLE: case BFA_FCPORT_SM_START: /* * Ignore event for a port as there is FAA misconfig */ break; case BFA_FCPORT_SM_DISABLE: if (bfa_fcport_send_disable(fcport)) bfa_sm_set_state(fcport, bfa_fcport_sm_disabling); else bfa_sm_set_state(fcport, bfa_fcport_sm_disabling_qwait); bfa_fcport_reset_linkinfo(fcport); bfa_fcport_scn(fcport, BFA_PORT_LINKDOWN, BFA_FALSE); bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_PORT_DISABLE, 0, "Port Disable"); bfa_fcport_aen_post(fcport, BFA_PORT_AEN_DISABLE); break; case BFA_FCPORT_SM_STOP: bfa_sm_set_state(fcport, bfa_fcport_sm_stopped); break; case BFA_FCPORT_SM_HWFAIL: bfa_fcport_reset_linkinfo(fcport); bfa_fcport_scn(fcport, BFA_PORT_LINKDOWN, BFA_FALSE); bfa_sm_set_state(fcport, bfa_fcport_sm_iocdown); break; default: bfa_sm_fault(fcport->bfa, event); } } /* * Link state is down */ static void bfa_fcport_ln_sm_dn(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKUP: bfa_sm_set_state(ln, bfa_fcport_ln_sm_up_nf); bfa_fcport_queue_cb(ln, BFA_PORT_LINKUP); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } /* * Link state is waiting for down notification */ static void bfa_fcport_ln_sm_dn_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKUP: bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn_up_nf); break; case BFA_FCPORT_LN_SM_NOTIFICATION: bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } /* * Link state is waiting for down notification and there is a pending up */ static void bfa_fcport_ln_sm_dn_up_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKDOWN: bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn_nf); break; case BFA_FCPORT_LN_SM_NOTIFICATION: bfa_sm_set_state(ln, bfa_fcport_ln_sm_up_nf); bfa_fcport_queue_cb(ln, BFA_PORT_LINKUP); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } /* * Link state is up */ static void bfa_fcport_ln_sm_up(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKDOWN: bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn_nf); bfa_fcport_queue_cb(ln, BFA_PORT_LINKDOWN); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } /* * Link state is waiting for up notification */ static void bfa_fcport_ln_sm_up_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKDOWN: bfa_sm_set_state(ln, bfa_fcport_ln_sm_up_dn_nf); break; case BFA_FCPORT_LN_SM_NOTIFICATION: bfa_sm_set_state(ln, bfa_fcport_ln_sm_up); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } /* * Link state is waiting for up notification and there is a pending down */ static void bfa_fcport_ln_sm_up_dn_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKUP: bfa_sm_set_state(ln, bfa_fcport_ln_sm_up_dn_up_nf); break; case BFA_FCPORT_LN_SM_NOTIFICATION: bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn_nf); bfa_fcport_queue_cb(ln, BFA_PORT_LINKDOWN); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } /* * Link state is waiting for up notification and there are pending down and up */ static void bfa_fcport_ln_sm_up_dn_up_nf(struct bfa_fcport_ln_s *ln, enum bfa_fcport_ln_sm_event event) { bfa_trc(ln->fcport->bfa, event); switch (event) { case BFA_FCPORT_LN_SM_LINKDOWN: bfa_sm_set_state(ln, bfa_fcport_ln_sm_up_dn_nf); break; case BFA_FCPORT_LN_SM_NOTIFICATION: bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn_up_nf); bfa_fcport_queue_cb(ln, BFA_PORT_LINKDOWN); break; default: bfa_sm_fault(ln->fcport->bfa, event); } } static void __bfa_cb_fcport_event(void *cbarg, bfa_boolean_t complete) { struct bfa_fcport_ln_s *ln = cbarg; if (complete) ln->fcport->event_cbfn(ln->fcport->event_cbarg, ln->ln_event); else bfa_sm_send_event(ln, BFA_FCPORT_LN_SM_NOTIFICATION); } /* * Send SCN notification to upper layers. * trunk - false if caller is fcport to ignore fcport event in trunked mode */ static void bfa_fcport_scn(struct bfa_fcport_s *fcport, enum bfa_port_linkstate event, bfa_boolean_t trunk) { if (fcport->cfg.trunked && !trunk) return; switch (event) { case BFA_PORT_LINKUP: bfa_sm_send_event(&fcport->ln, BFA_FCPORT_LN_SM_LINKUP); break; case BFA_PORT_LINKDOWN: bfa_sm_send_event(&fcport->ln, BFA_FCPORT_LN_SM_LINKDOWN); break; default: WARN_ON(1); } } static void bfa_fcport_queue_cb(struct bfa_fcport_ln_s *ln, enum bfa_port_linkstate event) { struct bfa_fcport_s *fcport = ln->fcport; if (fcport->bfa->fcs) { fcport->event_cbfn(fcport->event_cbarg, event); bfa_sm_send_event(ln, BFA_FCPORT_LN_SM_NOTIFICATION); } else { ln->ln_event = event; bfa_cb_queue(fcport->bfa, &ln->ln_qe, __bfa_cb_fcport_event, ln); } } #define FCPORT_STATS_DMA_SZ (BFA_ROUNDUP(sizeof(union bfa_fcport_stats_u), \ BFA_CACHELINE_SZ)) void bfa_fcport_meminfo(struct bfa_iocfc_cfg_s *cfg, struct bfa_meminfo_s *minfo, struct bfa_s *bfa) { struct bfa_mem_dma_s *fcport_dma = BFA_MEM_FCPORT_DMA(bfa); bfa_mem_dma_setup(minfo, fcport_dma, FCPORT_STATS_DMA_SZ); } static void bfa_fcport_qresume(void *cbarg) { struct bfa_fcport_s *fcport = cbarg; bfa_sm_send_event(fcport, BFA_FCPORT_SM_QRESUME); } static void bfa_fcport_mem_claim(struct bfa_fcport_s *fcport) { struct bfa_mem_dma_s *fcport_dma = &fcport->fcport_dma; fcport->stats_kva = bfa_mem_dma_virt(fcport_dma); fcport->stats_pa = bfa_mem_dma_phys(fcport_dma); fcport->stats = (union bfa_fcport_stats_u *) bfa_mem_dma_virt(fcport_dma); } /* * Memory initialization. */ void bfa_fcport_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); struct bfa_port_cfg_s *port_cfg = &fcport->cfg; struct bfa_fcport_ln_s *ln = &fcport->ln; struct timeval tv; fcport->bfa = bfa; ln->fcport = fcport; bfa_fcport_mem_claim(fcport); bfa_sm_set_state(fcport, bfa_fcport_sm_uninit); bfa_sm_set_state(ln, bfa_fcport_ln_sm_dn); /* * initialize time stamp for stats reset */ do_gettimeofday(&tv); fcport->stats_reset_time = tv.tv_sec; fcport->stats_dma_ready = BFA_FALSE; /* * initialize and set default configuration */ port_cfg->topology = BFA_PORT_TOPOLOGY_P2P; port_cfg->speed = BFA_PORT_SPEED_AUTO; port_cfg->trunked = BFA_FALSE; port_cfg->maxfrsize = 0; port_cfg->trl_def_speed = BFA_PORT_SPEED_1GBPS; port_cfg->qos_bw.high = BFA_QOS_BW_HIGH; port_cfg->qos_bw.med = BFA_QOS_BW_MED; port_cfg->qos_bw.low = BFA_QOS_BW_LOW; fcport->fec_state = BFA_FEC_OFFLINE; INIT_LIST_HEAD(&fcport->stats_pending_q); INIT_LIST_HEAD(&fcport->statsclr_pending_q); bfa_reqq_winit(&fcport->reqq_wait, bfa_fcport_qresume, fcport); } void bfa_fcport_start(struct bfa_s *bfa) { bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_START); } /* * Called when IOC failure is detected. */ void bfa_fcport_iocdisable(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_sm_send_event(fcport, BFA_FCPORT_SM_HWFAIL); bfa_trunk_iocdisable(bfa); } /* * Update loop info in fcport for SCN online */ static void bfa_fcport_update_loop_info(struct bfa_fcport_s *fcport, struct bfa_fcport_loop_info_s *loop_info) { fcport->myalpa = loop_info->myalpa; fcport->alpabm_valid = loop_info->alpabm_val; memcpy(fcport->alpabm.alpa_bm, loop_info->alpabm.alpa_bm, sizeof(struct fc_alpabm_s)); } static void bfa_fcport_update_linkinfo(struct bfa_fcport_s *fcport) { struct bfi_fcport_event_s *pevent = fcport->event_arg.i2hmsg.event; struct bfa_fcport_trunk_s *trunk = &fcport->trunk; fcport->speed = pevent->link_state.speed; fcport->topology = pevent->link_state.topology; if (fcport->topology == BFA_PORT_TOPOLOGY_LOOP) { bfa_fcport_update_loop_info(fcport, &pevent->link_state.attr.loop_info); return; } /* QoS Details */ fcport->qos_attr = pevent->link_state.qos_attr; fcport->qos_vc_attr = pevent->link_state.attr.vc_fcf.qos_vc_attr; if (fcport->cfg.bb_cr_enabled) fcport->bbcr_attr = pevent->link_state.attr.bbcr_attr; fcport->fec_state = pevent->link_state.fec_state; /* * update trunk state if applicable */ if (!fcport->cfg.trunked) trunk->attr.state = BFA_TRUNK_DISABLED; /* update FCoE specific */ fcport->fcoe_vlan = be16_to_cpu(pevent->link_state.attr.vc_fcf.fcf.vlan); bfa_trc(fcport->bfa, fcport->speed); bfa_trc(fcport->bfa, fcport->topology); } static void bfa_fcport_reset_linkinfo(struct bfa_fcport_s *fcport) { fcport->speed = BFA_PORT_SPEED_UNKNOWN; fcport->topology = BFA_PORT_TOPOLOGY_NONE; fcport->fec_state = BFA_FEC_OFFLINE; } /* * Send port enable message to firmware. */ static bfa_boolean_t bfa_fcport_send_enable(struct bfa_fcport_s *fcport) { struct bfi_fcport_enable_req_s *m; /* * Increment message tag before queue check, so that responses to old * requests are discarded. */ fcport->msgtag++; /* * check for room in queue to send request now */ m = bfa_reqq_next(fcport->bfa, BFA_REQQ_PORT); if (!m) { bfa_reqq_wait(fcport->bfa, BFA_REQQ_PORT, &fcport->reqq_wait); return BFA_FALSE; } bfi_h2i_set(m->mh, BFI_MC_FCPORT, BFI_FCPORT_H2I_ENABLE_REQ, bfa_fn_lpu(fcport->bfa)); m->nwwn = fcport->nwwn; m->pwwn = fcport->pwwn; m->port_cfg = fcport->cfg; m->msgtag = fcport->msgtag; m->port_cfg.maxfrsize = cpu_to_be16(fcport->cfg.maxfrsize); m->use_flash_cfg = fcport->use_flash_cfg; bfa_dma_be_addr_set(m->stats_dma_addr, fcport->stats_pa); bfa_trc(fcport->bfa, m->stats_dma_addr.a32.addr_lo); bfa_trc(fcport->bfa, m->stats_dma_addr.a32.addr_hi); /* * queue I/O message to firmware */ bfa_reqq_produce(fcport->bfa, BFA_REQQ_PORT, m->mh); return BFA_TRUE; } /* * Send port disable message to firmware. */ static bfa_boolean_t bfa_fcport_send_disable(struct bfa_fcport_s *fcport) { struct bfi_fcport_req_s *m; /* * Increment message tag before queue check, so that responses to old * requests are discarded. */ fcport->msgtag++; /* * check for room in queue to send request now */ m = bfa_reqq_next(fcport->bfa, BFA_REQQ_PORT); if (!m) { bfa_reqq_wait(fcport->bfa, BFA_REQQ_PORT, &fcport->reqq_wait); return BFA_FALSE; } bfi_h2i_set(m->mh, BFI_MC_FCPORT, BFI_FCPORT_H2I_DISABLE_REQ, bfa_fn_lpu(fcport->bfa)); m->msgtag = fcport->msgtag; /* * queue I/O message to firmware */ bfa_reqq_produce(fcport->bfa, BFA_REQQ_PORT, m->mh); return BFA_TRUE; } static void bfa_fcport_set_wwns(struct bfa_fcport_s *fcport) { fcport->pwwn = fcport->bfa->ioc.attr->pwwn; fcport->nwwn = fcport->bfa->ioc.attr->nwwn; bfa_trc(fcport->bfa, fcport->pwwn); bfa_trc(fcport->bfa, fcport->nwwn); } static void bfa_fcport_qos_stats_swap(struct bfa_qos_stats_s *d, struct bfa_qos_stats_s *s) { u32 *dip = (u32 *) d; __be32 *sip = (__be32 *) s; int i; /* Now swap the 32 bit fields */ for (i = 0; i < (sizeof(struct bfa_qos_stats_s)/sizeof(u32)); ++i) dip[i] = be32_to_cpu(sip[i]); } static void bfa_fcport_fcoe_stats_swap(struct bfa_fcoe_stats_s *d, struct bfa_fcoe_stats_s *s) { u32 *dip = (u32 *) d; __be32 *sip = (__be32 *) s; int i; for (i = 0; i < ((sizeof(struct bfa_fcoe_stats_s))/sizeof(u32)); i = i + 2) { #ifdef __BIG_ENDIAN dip[i] = be32_to_cpu(sip[i]); dip[i + 1] = be32_to_cpu(sip[i + 1]); #else dip[i] = be32_to_cpu(sip[i + 1]); dip[i + 1] = be32_to_cpu(sip[i]); #endif } } static void __bfa_cb_fcport_stats_get(void *cbarg, bfa_boolean_t complete) { struct bfa_fcport_s *fcport = (struct bfa_fcport_s *)cbarg; struct bfa_cb_pending_q_s *cb; struct list_head *qe, *qen; union bfa_fcport_stats_u *ret; if (complete) { struct timeval tv; if (fcport->stats_status == BFA_STATUS_OK) do_gettimeofday(&tv); list_for_each_safe(qe, qen, &fcport->stats_pending_q) { bfa_q_deq(&fcport->stats_pending_q, &qe); cb = (struct bfa_cb_pending_q_s *)qe; if (fcport->stats_status == BFA_STATUS_OK) { ret = (union bfa_fcport_stats_u *)cb->data; /* Swap FC QoS or FCoE stats */ if (bfa_ioc_get_fcmode(&fcport->bfa->ioc)) bfa_fcport_qos_stats_swap(&ret->fcqos, &fcport->stats->fcqos); else { bfa_fcport_fcoe_stats_swap(&ret->fcoe, &fcport->stats->fcoe); ret->fcoe.secs_reset = tv.tv_sec - fcport->stats_reset_time; } } bfa_cb_queue_status(fcport->bfa, &cb->hcb_qe, fcport->stats_status); } fcport->stats_status = BFA_STATUS_OK; } else { INIT_LIST_HEAD(&fcport->stats_pending_q); fcport->stats_status = BFA_STATUS_OK; } } static void bfa_fcport_stats_get_timeout(void *cbarg) { struct bfa_fcport_s *fcport = (struct bfa_fcport_s *) cbarg; bfa_trc(fcport->bfa, fcport->stats_qfull); if (fcport->stats_qfull) { bfa_reqq_wcancel(&fcport->stats_reqq_wait); fcport->stats_qfull = BFA_FALSE; } fcport->stats_status = BFA_STATUS_ETIMER; __bfa_cb_fcport_stats_get(fcport, BFA_TRUE); } static void bfa_fcport_send_stats_get(void *cbarg) { struct bfa_fcport_s *fcport = (struct bfa_fcport_s *) cbarg; struct bfi_fcport_req_s *msg; msg = bfa_reqq_next(fcport->bfa, BFA_REQQ_PORT); if (!msg) { fcport->stats_qfull = BFA_TRUE; bfa_reqq_winit(&fcport->stats_reqq_wait, bfa_fcport_send_stats_get, fcport); bfa_reqq_wait(fcport->bfa, BFA_REQQ_PORT, &fcport->stats_reqq_wait); return; } fcport->stats_qfull = BFA_FALSE; memset(msg, 0, sizeof(struct bfi_fcport_req_s)); bfi_h2i_set(msg->mh, BFI_MC_FCPORT, BFI_FCPORT_H2I_STATS_GET_REQ, bfa_fn_lpu(fcport->bfa)); bfa_reqq_produce(fcport->bfa, BFA_REQQ_PORT, msg->mh); } static void __bfa_cb_fcport_stats_clr(void *cbarg, bfa_boolean_t complete) { struct bfa_fcport_s *fcport = (struct bfa_fcport_s *) cbarg; struct bfa_cb_pending_q_s *cb; struct list_head *qe, *qen; if (complete) { struct timeval tv; /* * re-initialize time stamp for stats reset */ do_gettimeofday(&tv); fcport->stats_reset_time = tv.tv_sec; list_for_each_safe(qe, qen, &fcport->statsclr_pending_q) { bfa_q_deq(&fcport->statsclr_pending_q, &qe); cb = (struct bfa_cb_pending_q_s *)qe; bfa_cb_queue_status(fcport->bfa, &cb->hcb_qe, fcport->stats_status); } fcport->stats_status = BFA_STATUS_OK; } else { INIT_LIST_HEAD(&fcport->statsclr_pending_q); fcport->stats_status = BFA_STATUS_OK; } } static void bfa_fcport_stats_clr_timeout(void *cbarg) { struct bfa_fcport_s *fcport = (struct bfa_fcport_s *) cbarg; bfa_trc(fcport->bfa, fcport->stats_qfull); if (fcport->stats_qfull) { bfa_reqq_wcancel(&fcport->stats_reqq_wait); fcport->stats_qfull = BFA_FALSE; } fcport->stats_status = BFA_STATUS_ETIMER; __bfa_cb_fcport_stats_clr(fcport, BFA_TRUE); } static void bfa_fcport_send_stats_clear(void *cbarg) { struct bfa_fcport_s *fcport = (struct bfa_fcport_s *) cbarg; struct bfi_fcport_req_s *msg; msg = bfa_reqq_next(fcport->bfa, BFA_REQQ_PORT); if (!msg) { fcport->stats_qfull = BFA_TRUE; bfa_reqq_winit(&fcport->stats_reqq_wait, bfa_fcport_send_stats_clear, fcport); bfa_reqq_wait(fcport->bfa, BFA_REQQ_PORT, &fcport->stats_reqq_wait); return; } fcport->stats_qfull = BFA_FALSE; memset(msg, 0, sizeof(struct bfi_fcport_req_s)); bfi_h2i_set(msg->mh, BFI_MC_FCPORT, BFI_FCPORT_H2I_STATS_CLEAR_REQ, bfa_fn_lpu(fcport->bfa)); bfa_reqq_produce(fcport->bfa, BFA_REQQ_PORT, msg->mh); } /* * Handle trunk SCN event from firmware. */ static void bfa_trunk_scn(struct bfa_fcport_s *fcport, struct bfi_fcport_trunk_scn_s *scn) { struct bfa_fcport_trunk_s *trunk = &fcport->trunk; struct bfi_fcport_trunk_link_s *tlink; struct bfa_trunk_link_attr_s *lattr; enum bfa_trunk_state state_prev; int i; int link_bm = 0; bfa_trc(fcport->bfa, fcport->cfg.trunked); WARN_ON(scn->trunk_state != BFA_TRUNK_ONLINE && scn->trunk_state != BFA_TRUNK_OFFLINE); bfa_trc(fcport->bfa, trunk->attr.state); bfa_trc(fcport->bfa, scn->trunk_state); bfa_trc(fcport->bfa, scn->trunk_speed); /* * Save off new state for trunk attribute query */ state_prev = trunk->attr.state; if (fcport->cfg.trunked && (trunk->attr.state != BFA_TRUNK_DISABLED)) trunk->attr.state = scn->trunk_state; trunk->attr.speed = scn->trunk_speed; for (i = 0; i < BFA_TRUNK_MAX_PORTS; i++) { lattr = &trunk->attr.link_attr[i]; tlink = &scn->tlink[i]; lattr->link_state = tlink->state; lattr->trunk_wwn = tlink->trunk_wwn; lattr->fctl = tlink->fctl; lattr->speed = tlink->speed; lattr->deskew = be32_to_cpu(tlink->deskew); if (tlink->state == BFA_TRUNK_LINK_STATE_UP) { fcport->speed = tlink->speed; fcport->topology = BFA_PORT_TOPOLOGY_P2P; link_bm |= 1 << i; } bfa_trc(fcport->bfa, lattr->link_state); bfa_trc(fcport->bfa, lattr->trunk_wwn); bfa_trc(fcport->bfa, lattr->fctl); bfa_trc(fcport->bfa, lattr->speed); bfa_trc(fcport->bfa, lattr->deskew); } switch (link_bm) { case 3: bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_TRUNK_SCN, 0, "Trunk up(0,1)"); break; case 2: bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_TRUNK_SCN, 0, "Trunk up(-,1)"); break; case 1: bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_TRUNK_SCN, 0, "Trunk up(0,-)"); break; default: bfa_plog_str(fcport->bfa->plog, BFA_PL_MID_HAL, BFA_PL_EID_TRUNK_SCN, 0, "Trunk down"); } /* * Notify upper layers if trunk state changed. */ if ((state_prev != trunk->attr.state) || (scn->trunk_state == BFA_TRUNK_OFFLINE)) { bfa_fcport_scn(fcport, (scn->trunk_state == BFA_TRUNK_ONLINE) ? BFA_PORT_LINKUP : BFA_PORT_LINKDOWN, BFA_TRUE); } } static void bfa_trunk_iocdisable(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); int i = 0; /* * In trunked mode, notify upper layers that link is down */ if (fcport->cfg.trunked) { if (fcport->trunk.attr.state == BFA_TRUNK_ONLINE) bfa_fcport_scn(fcport, BFA_PORT_LINKDOWN, BFA_TRUE); fcport->trunk.attr.state = BFA_TRUNK_OFFLINE; fcport->trunk.attr.speed = BFA_PORT_SPEED_UNKNOWN; for (i = 0; i < BFA_TRUNK_MAX_PORTS; i++) { fcport->trunk.attr.link_attr[i].trunk_wwn = 0; fcport->trunk.attr.link_attr[i].fctl = BFA_TRUNK_LINK_FCTL_NORMAL; fcport->trunk.attr.link_attr[i].link_state = BFA_TRUNK_LINK_STATE_DN_LINKDN; fcport->trunk.attr.link_attr[i].speed = BFA_PORT_SPEED_UNKNOWN; fcport->trunk.attr.link_attr[i].deskew = 0; } } } /* * Called to initialize port attributes */ void bfa_fcport_init(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); /* * Initialize port attributes from IOC hardware data. */ bfa_fcport_set_wwns(fcport); if (fcport->cfg.maxfrsize == 0) fcport->cfg.maxfrsize = bfa_ioc_maxfrsize(&bfa->ioc); fcport->cfg.rx_bbcredit = bfa_ioc_rx_bbcredit(&bfa->ioc); fcport->speed_sup = bfa_ioc_speed_sup(&bfa->ioc); if (bfa_fcport_is_pbcdisabled(bfa)) bfa->modules.port.pbc_disabled = BFA_TRUE; WARN_ON(!fcport->cfg.maxfrsize); WARN_ON(!fcport->cfg.rx_bbcredit); WARN_ON(!fcport->speed_sup); } /* * Firmware message handler. */ void bfa_fcport_isr(struct bfa_s *bfa, struct bfi_msg_s *msg) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); union bfi_fcport_i2h_msg_u i2hmsg; i2hmsg.msg = msg; fcport->event_arg.i2hmsg = i2hmsg; bfa_trc(bfa, msg->mhdr.msg_id); bfa_trc(bfa, bfa_sm_to_state(hal_port_sm_table, fcport->sm)); switch (msg->mhdr.msg_id) { case BFI_FCPORT_I2H_ENABLE_RSP: if (fcport->msgtag == i2hmsg.penable_rsp->msgtag) { fcport->stats_dma_ready = BFA_TRUE; if (fcport->use_flash_cfg) { fcport->cfg = i2hmsg.penable_rsp->port_cfg; fcport->cfg.maxfrsize = cpu_to_be16(fcport->cfg.maxfrsize); fcport->cfg.path_tov = cpu_to_be16(fcport->cfg.path_tov); fcport->cfg.q_depth = cpu_to_be16(fcport->cfg.q_depth); if (fcport->cfg.trunked) fcport->trunk.attr.state = BFA_TRUNK_OFFLINE; else fcport->trunk.attr.state = BFA_TRUNK_DISABLED; fcport->qos_attr.qos_bw = i2hmsg.penable_rsp->port_cfg.qos_bw; fcport->use_flash_cfg = BFA_FALSE; } if (fcport->cfg.qos_enabled) fcport->qos_attr.state = BFA_QOS_OFFLINE; else fcport->qos_attr.state = BFA_QOS_DISABLED; fcport->qos_attr.qos_bw_op = i2hmsg.penable_rsp->port_cfg.qos_bw; if (fcport->cfg.bb_cr_enabled) fcport->bbcr_attr.state = BFA_BBCR_OFFLINE; else fcport->bbcr_attr.state = BFA_BBCR_DISABLED; bfa_sm_send_event(fcport, BFA_FCPORT_SM_FWRSP); } break; case BFI_FCPORT_I2H_DISABLE_RSP: if (fcport->msgtag == i2hmsg.penable_rsp->msgtag) bfa_sm_send_event(fcport, BFA_FCPORT_SM_FWRSP); break; case BFI_FCPORT_I2H_EVENT: if (fcport->cfg.bb_cr_enabled) fcport->bbcr_attr.state = BFA_BBCR_OFFLINE; else fcport->bbcr_attr.state = BFA_BBCR_DISABLED; if (i2hmsg.event->link_state.linkstate == BFA_PORT_LINKUP) bfa_sm_send_event(fcport, BFA_FCPORT_SM_LINKUP); else { if (i2hmsg.event->link_state.linkstate_rsn == BFA_PORT_LINKSTATE_RSN_FAA_MISCONFIG) bfa_sm_send_event(fcport, BFA_FCPORT_SM_FAA_MISCONFIG); else bfa_sm_send_event(fcport, BFA_FCPORT_SM_LINKDOWN); } fcport->qos_attr.qos_bw_op = i2hmsg.event->link_state.qos_attr.qos_bw_op; break; case BFI_FCPORT_I2H_TRUNK_SCN: bfa_trunk_scn(fcport, i2hmsg.trunk_scn); break; case BFI_FCPORT_I2H_STATS_GET_RSP: /* * check for timer pop before processing the rsp */ if (list_empty(&fcport->stats_pending_q) || (fcport->stats_status == BFA_STATUS_ETIMER)) break; bfa_timer_stop(&fcport->timer); fcport->stats_status = i2hmsg.pstatsget_rsp->status; __bfa_cb_fcport_stats_get(fcport, BFA_TRUE); break; case BFI_FCPORT_I2H_STATS_CLEAR_RSP: /* * check for timer pop before processing the rsp */ if (list_empty(&fcport->statsclr_pending_q) || (fcport->stats_status == BFA_STATUS_ETIMER)) break; bfa_timer_stop(&fcport->timer); fcport->stats_status = BFA_STATUS_OK; __bfa_cb_fcport_stats_clr(fcport, BFA_TRUE); break; case BFI_FCPORT_I2H_ENABLE_AEN: bfa_sm_send_event(fcport, BFA_FCPORT_SM_ENABLE); break; case BFI_FCPORT_I2H_DISABLE_AEN: bfa_sm_send_event(fcport, BFA_FCPORT_SM_DISABLE); break; default: WARN_ON(1); break; } } /* * Registered callback for port events. */ void bfa_fcport_event_register(struct bfa_s *bfa, void (*cbfn) (void *cbarg, enum bfa_port_linkstate event), void *cbarg) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); fcport->event_cbfn = cbfn; fcport->event_cbarg = cbarg; } bfa_status_t bfa_fcport_enable(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); if (bfa_fcport_is_pbcdisabled(bfa)) return BFA_STATUS_PBC; if (bfa_ioc_is_disabled(&bfa->ioc)) return BFA_STATUS_IOC_DISABLED; if (fcport->diag_busy) return BFA_STATUS_DIAG_BUSY; bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_ENABLE); return BFA_STATUS_OK; } bfa_status_t bfa_fcport_disable(struct bfa_s *bfa) { if (bfa_fcport_is_pbcdisabled(bfa)) return BFA_STATUS_PBC; if (bfa_ioc_is_disabled(&bfa->ioc)) return BFA_STATUS_IOC_DISABLED; bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_DISABLE); return BFA_STATUS_OK; } /* If PBC is disabled on port, return error */ bfa_status_t bfa_fcport_is_pbcdisabled(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); struct bfa_iocfc_s *iocfc = &bfa->iocfc; struct bfi_iocfc_cfgrsp_s *cfgrsp = iocfc->cfgrsp; if (cfgrsp->pbc_cfg.port_enabled == BFI_PBC_PORT_DISABLED) { bfa_trc(bfa, fcport->pwwn); return BFA_STATUS_PBC; } return BFA_STATUS_OK; } /* * Configure port speed. */ bfa_status_t bfa_fcport_cfg_speed(struct bfa_s *bfa, enum bfa_port_speed speed) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, speed); if (fcport->cfg.trunked == BFA_TRUE) return BFA_STATUS_TRUNK_ENABLED; if ((fcport->cfg.topology == BFA_PORT_TOPOLOGY_LOOP) && (speed == BFA_PORT_SPEED_16GBPS)) return BFA_STATUS_UNSUPP_SPEED; if ((speed != BFA_PORT_SPEED_AUTO) && (speed > fcport->speed_sup)) { bfa_trc(bfa, fcport->speed_sup); return BFA_STATUS_UNSUPP_SPEED; } /* Port speed entered needs to be checked */ if (bfa_ioc_get_type(&fcport->bfa->ioc) == BFA_IOC_TYPE_FC) { /* For CT2, 1G is not supported */ if ((speed == BFA_PORT_SPEED_1GBPS) && (bfa_asic_id_ct2(bfa->ioc.pcidev.device_id))) return BFA_STATUS_UNSUPP_SPEED; /* Already checked for Auto Speed and Max Speed supp */ if (!(speed == BFA_PORT_SPEED_1GBPS || speed == BFA_PORT_SPEED_2GBPS || speed == BFA_PORT_SPEED_4GBPS || speed == BFA_PORT_SPEED_8GBPS || speed == BFA_PORT_SPEED_16GBPS || speed == BFA_PORT_SPEED_AUTO)) return BFA_STATUS_UNSUPP_SPEED; } else { if (speed != BFA_PORT_SPEED_10GBPS) return BFA_STATUS_UNSUPP_SPEED; } fcport->cfg.speed = speed; return BFA_STATUS_OK; } /* * Get current speed. */ enum bfa_port_speed bfa_fcport_get_speed(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->speed; } /* * Configure port topology. */ bfa_status_t bfa_fcport_cfg_topology(struct bfa_s *bfa, enum bfa_port_topology topology) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, topology); bfa_trc(bfa, fcport->cfg.topology); switch (topology) { case BFA_PORT_TOPOLOGY_P2P: break; case BFA_PORT_TOPOLOGY_LOOP: if ((bfa_fcport_is_qos_enabled(bfa) != BFA_FALSE) || (fcport->qos_attr.state != BFA_QOS_DISABLED)) return BFA_STATUS_ERROR_QOS_ENABLED; if (fcport->cfg.ratelimit != BFA_FALSE) return BFA_STATUS_ERROR_TRL_ENABLED; if ((bfa_fcport_is_trunk_enabled(bfa) != BFA_FALSE) || (fcport->trunk.attr.state != BFA_TRUNK_DISABLED)) return BFA_STATUS_ERROR_TRUNK_ENABLED; if ((bfa_fcport_get_speed(bfa) == BFA_PORT_SPEED_16GBPS) || (fcport->cfg.speed == BFA_PORT_SPEED_16GBPS)) return BFA_STATUS_UNSUPP_SPEED; if (bfa_mfg_is_mezz(bfa->ioc.attr->card_type)) return BFA_STATUS_LOOP_UNSUPP_MEZZ; if (bfa_fcport_is_dport(bfa) != BFA_FALSE) return BFA_STATUS_DPORT_ERR; if (bfa_fcport_is_ddport(bfa) != BFA_FALSE) return BFA_STATUS_DPORT_ERR; break; case BFA_PORT_TOPOLOGY_AUTO: break; default: return BFA_STATUS_EINVAL; } fcport->cfg.topology = topology; return BFA_STATUS_OK; } /* * Get current topology. */ enum bfa_port_topology bfa_fcport_get_topology(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->topology; } /** * Get config topology. */ enum bfa_port_topology bfa_fcport_get_cfg_topology(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->cfg.topology; } bfa_status_t bfa_fcport_cfg_hardalpa(struct bfa_s *bfa, u8 alpa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, alpa); bfa_trc(bfa, fcport->cfg.cfg_hardalpa); bfa_trc(bfa, fcport->cfg.hardalpa); fcport->cfg.cfg_hardalpa = BFA_TRUE; fcport->cfg.hardalpa = alpa; return BFA_STATUS_OK; } bfa_status_t bfa_fcport_clr_hardalpa(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, fcport->cfg.cfg_hardalpa); bfa_trc(bfa, fcport->cfg.hardalpa); fcport->cfg.cfg_hardalpa = BFA_FALSE; return BFA_STATUS_OK; } bfa_boolean_t bfa_fcport_get_hardalpa(struct bfa_s *bfa, u8 *alpa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); *alpa = fcport->cfg.hardalpa; return fcport->cfg.cfg_hardalpa; } u8 bfa_fcport_get_myalpa(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->myalpa; } bfa_status_t bfa_fcport_cfg_maxfrsize(struct bfa_s *bfa, u16 maxfrsize) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, maxfrsize); bfa_trc(bfa, fcport->cfg.maxfrsize); /* with in range */ if ((maxfrsize > FC_MAX_PDUSZ) || (maxfrsize < FC_MIN_PDUSZ)) return BFA_STATUS_INVLD_DFSZ; /* power of 2, if not the max frame size of 2112 */ if ((maxfrsize != FC_MAX_PDUSZ) && (maxfrsize & (maxfrsize - 1))) return BFA_STATUS_INVLD_DFSZ; fcport->cfg.maxfrsize = maxfrsize; return BFA_STATUS_OK; } u16 bfa_fcport_get_maxfrsize(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->cfg.maxfrsize; } u8 bfa_fcport_get_rx_bbcredit(struct bfa_s *bfa) { if (bfa_fcport_get_topology(bfa) != BFA_PORT_TOPOLOGY_LOOP) return (BFA_FCPORT_MOD(bfa))->cfg.rx_bbcredit; else return 0; } void bfa_fcport_set_tx_bbcredit(struct bfa_s *bfa, u16 tx_bbcredit) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); fcport->cfg.tx_bbcredit = (u8)tx_bbcredit; } /* * Get port attributes. */ wwn_t bfa_fcport_get_wwn(struct bfa_s *bfa, bfa_boolean_t node) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); if (node) return fcport->nwwn; else return fcport->pwwn; } void bfa_fcport_get_attr(struct bfa_s *bfa, struct bfa_port_attr_s *attr) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); memset(attr, 0, sizeof(struct bfa_port_attr_s)); attr->nwwn = fcport->nwwn; attr->pwwn = fcport->pwwn; attr->factorypwwn = bfa->ioc.attr->mfg_pwwn; attr->factorynwwn = bfa->ioc.attr->mfg_nwwn; memcpy(&attr->pport_cfg, &fcport->cfg, sizeof(struct bfa_port_cfg_s)); /* speed attributes */ attr->pport_cfg.speed = fcport->cfg.speed; attr->speed_supported = fcport->speed_sup; attr->speed = fcport->speed; attr->cos_supported = FC_CLASS_3; /* topology attributes */ attr->pport_cfg.topology = fcport->cfg.topology; attr->topology = fcport->topology; attr->pport_cfg.trunked = fcport->cfg.trunked; /* beacon attributes */ attr->beacon = fcport->beacon; attr->link_e2e_beacon = fcport->link_e2e_beacon; attr->pport_cfg.path_tov = bfa_fcpim_path_tov_get(bfa); attr->pport_cfg.q_depth = bfa_fcpim_qdepth_get(bfa); attr->port_state = bfa_sm_to_state(hal_port_sm_table, fcport->sm); attr->fec_state = fcport->fec_state; /* PBC Disabled State */ if (bfa_fcport_is_pbcdisabled(bfa)) attr->port_state = BFA_PORT_ST_PREBOOT_DISABLED; else { if (bfa_ioc_is_disabled(&fcport->bfa->ioc)) attr->port_state = BFA_PORT_ST_IOCDIS; else if (bfa_ioc_fw_mismatch(&fcport->bfa->ioc)) attr->port_state = BFA_PORT_ST_FWMISMATCH; } /* FCoE vlan */ attr->fcoe_vlan = fcport->fcoe_vlan; } #define BFA_FCPORT_STATS_TOV 1000 /* * Fetch port statistics (FCQoS or FCoE). */ bfa_status_t bfa_fcport_get_stats(struct bfa_s *bfa, struct bfa_cb_pending_q_s *cb) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); if (!bfa_iocfc_is_operational(bfa) || !fcport->stats_dma_ready) return BFA_STATUS_IOC_NON_OP; if (!list_empty(&fcport->statsclr_pending_q)) return BFA_STATUS_DEVBUSY; if (list_empty(&fcport->stats_pending_q)) { list_add_tail(&cb->hcb_qe.qe, &fcport->stats_pending_q); bfa_fcport_send_stats_get(fcport); bfa_timer_start(bfa, &fcport->timer, bfa_fcport_stats_get_timeout, fcport, BFA_FCPORT_STATS_TOV); } else list_add_tail(&cb->hcb_qe.qe, &fcport->stats_pending_q); return BFA_STATUS_OK; } /* * Reset port statistics (FCQoS or FCoE). */ bfa_status_t bfa_fcport_clear_stats(struct bfa_s *bfa, struct bfa_cb_pending_q_s *cb) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); if (!bfa_iocfc_is_operational(bfa) || !fcport->stats_dma_ready) return BFA_STATUS_IOC_NON_OP; if (!list_empty(&fcport->stats_pending_q)) return BFA_STATUS_DEVBUSY; if (list_empty(&fcport->statsclr_pending_q)) { list_add_tail(&cb->hcb_qe.qe, &fcport->statsclr_pending_q); bfa_fcport_send_stats_clear(fcport); bfa_timer_start(bfa, &fcport->timer, bfa_fcport_stats_clr_timeout, fcport, BFA_FCPORT_STATS_TOV); } else list_add_tail(&cb->hcb_qe.qe, &fcport->statsclr_pending_q); return BFA_STATUS_OK; } /* * Fetch port attributes. */ bfa_boolean_t bfa_fcport_is_disabled(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return bfa_sm_to_state(hal_port_sm_table, fcport->sm) == BFA_PORT_ST_DISABLED; } bfa_boolean_t bfa_fcport_is_dport(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return (bfa_sm_to_state(hal_port_sm_table, fcport->sm) == BFA_PORT_ST_DPORT); } bfa_boolean_t bfa_fcport_is_ddport(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return (bfa_sm_to_state(hal_port_sm_table, fcport->sm) == BFA_PORT_ST_DDPORT); } bfa_status_t bfa_fcport_set_qos_bw(struct bfa_s *bfa, struct bfa_qos_bw_s *qos_bw) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); enum bfa_ioc_type_e ioc_type = bfa_get_type(bfa); bfa_trc(bfa, ioc_type); if ((qos_bw->high == 0) || (qos_bw->med == 0) || (qos_bw->low == 0)) return BFA_STATUS_QOS_BW_INVALID; if ((qos_bw->high + qos_bw->med + qos_bw->low) != 100) return BFA_STATUS_QOS_BW_INVALID; if ((qos_bw->med > qos_bw->high) || (qos_bw->low > qos_bw->med) || (qos_bw->low > qos_bw->high)) return BFA_STATUS_QOS_BW_INVALID; if ((ioc_type == BFA_IOC_TYPE_FC) && (fcport->cfg.topology != BFA_PORT_TOPOLOGY_LOOP)) fcport->cfg.qos_bw = *qos_bw; return BFA_STATUS_OK; } bfa_boolean_t bfa_fcport_is_ratelim(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->cfg.ratelimit ? BFA_TRUE : BFA_FALSE; } /* * Enable/Disable FAA feature in port config */ void bfa_fcport_cfg_faa(struct bfa_s *bfa, u8 state) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, state); fcport->cfg.faa_state = state; } /* * Get default minimum ratelim speed */ enum bfa_port_speed bfa_fcport_get_ratelim_speed(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, fcport->cfg.trl_def_speed); return fcport->cfg.trl_def_speed; } void bfa_fcport_beacon(void *dev, bfa_boolean_t beacon, bfa_boolean_t link_e2e_beacon) { struct bfa_s *bfa = dev; struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, beacon); bfa_trc(bfa, link_e2e_beacon); bfa_trc(bfa, fcport->beacon); bfa_trc(bfa, fcport->link_e2e_beacon); fcport->beacon = beacon; fcport->link_e2e_beacon = link_e2e_beacon; } bfa_boolean_t bfa_fcport_is_linkup(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return (!fcport->cfg.trunked && bfa_sm_cmp_state(fcport, bfa_fcport_sm_linkup)) || (fcport->cfg.trunked && fcport->trunk.attr.state == BFA_TRUNK_ONLINE); } bfa_boolean_t bfa_fcport_is_qos_enabled(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->cfg.qos_enabled; } bfa_boolean_t bfa_fcport_is_trunk_enabled(struct bfa_s *bfa) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); return fcport->cfg.trunked; } bfa_status_t bfa_fcport_cfg_bbcr(struct bfa_s *bfa, bfa_boolean_t on_off, u8 bb_scn) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); bfa_trc(bfa, on_off); if (bfa_ioc_get_type(&fcport->bfa->ioc) != BFA_IOC_TYPE_FC) return BFA_STATUS_BBCR_FC_ONLY; if (bfa_mfg_is_mezz(bfa->ioc.attr->card_type) && (bfa->ioc.attr->card_type != BFA_MFG_TYPE_CHINOOK)) return BFA_STATUS_CMD_NOTSUPP_MEZZ; if (on_off) { if (fcport->cfg.topology == BFA_PORT_TOPOLOGY_LOOP) return BFA_STATUS_TOPOLOGY_LOOP; if (fcport->cfg.qos_enabled) return BFA_STATUS_ERROR_QOS_ENABLED; if (fcport->cfg.trunked) return BFA_STATUS_TRUNK_ENABLED; if ((fcport->cfg.speed != BFA_PORT_SPEED_AUTO) && (fcport->cfg.speed < bfa_ioc_speed_sup(&bfa->ioc))) return BFA_STATUS_ERR_BBCR_SPEED_UNSUPPORT; if (bfa_ioc_speed_sup(&bfa->ioc) < BFA_PORT_SPEED_8GBPS) return BFA_STATUS_FEATURE_NOT_SUPPORTED; if (fcport->cfg.bb_cr_enabled) { if (bb_scn != fcport->cfg.bb_scn) return BFA_STATUS_BBCR_CFG_NO_CHANGE; else return BFA_STATUS_NO_CHANGE; } if ((bb_scn == 0) || (bb_scn > BFA_BB_SCN_MAX)) bb_scn = BFA_BB_SCN_DEF; fcport->cfg.bb_cr_enabled = on_off; fcport->cfg.bb_scn = bb_scn; } else { if (!fcport->cfg.bb_cr_enabled) return BFA_STATUS_NO_CHANGE; fcport->cfg.bb_cr_enabled = on_off; fcport->cfg.bb_scn = 0; } return BFA_STATUS_OK; } bfa_status_t bfa_fcport_get_bbcr_attr(struct bfa_s *bfa, struct bfa_bbcr_attr_s *bbcr_attr) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); if (bfa_ioc_get_type(&fcport->bfa->ioc) != BFA_IOC_TYPE_FC) return BFA_STATUS_BBCR_FC_ONLY; if (fcport->cfg.topology == BFA_PORT_TOPOLOGY_LOOP) return BFA_STATUS_TOPOLOGY_LOOP; *bbcr_attr = fcport->bbcr_attr; return BFA_STATUS_OK; } void bfa_fcport_dportenable(struct bfa_s *bfa) { /* * Assume caller check for port is in disable state */ bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_DPORTENABLE); bfa_port_set_dportenabled(&bfa->modules.port, BFA_TRUE); } void bfa_fcport_dportdisable(struct bfa_s *bfa) { /* * Assume caller check for port is in disable state */ bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_DPORTDISABLE); bfa_port_set_dportenabled(&bfa->modules.port, BFA_FALSE); } void bfa_fcport_ddportenable(struct bfa_s *bfa) { /* * Assume caller check for port is in disable state */ bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_DDPORTENABLE); } void bfa_fcport_ddportdisable(struct bfa_s *bfa) { /* * Assume caller check for port is in disable state */ bfa_sm_send_event(BFA_FCPORT_MOD(bfa), BFA_FCPORT_SM_DDPORTDISABLE); } /* * Rport State machine functions */ /* * Beginning state, only online event expected. */ static void bfa_rport_sm_uninit(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_CREATE: bfa_stats(rp, sm_un_cr); bfa_sm_set_state(rp, bfa_rport_sm_created); break; default: bfa_stats(rp, sm_un_unexp); bfa_sm_fault(rp->bfa, event); } } static void bfa_rport_sm_created(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_ONLINE: bfa_stats(rp, sm_cr_on); if (bfa_rport_send_fwcreate(rp)) bfa_sm_set_state(rp, bfa_rport_sm_fwcreate); else bfa_sm_set_state(rp, bfa_rport_sm_fwcreate_qfull); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_cr_del); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_rport_free(rp); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_cr_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); break; default: bfa_stats(rp, sm_cr_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Waiting for rport create response from firmware. */ static void bfa_rport_sm_fwcreate(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_FWRSP: bfa_stats(rp, sm_fwc_rsp); bfa_sm_set_state(rp, bfa_rport_sm_online); bfa_rport_online_cb(rp); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_fwc_del); bfa_sm_set_state(rp, bfa_rport_sm_delete_pending); break; case BFA_RPORT_SM_OFFLINE: bfa_stats(rp, sm_fwc_off); bfa_sm_set_state(rp, bfa_rport_sm_offline_pending); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_fwc_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); break; default: bfa_stats(rp, sm_fwc_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Request queue is full, awaiting queue resume to send create request. */ static void bfa_rport_sm_fwcreate_qfull(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_QRESUME: bfa_sm_set_state(rp, bfa_rport_sm_fwcreate); bfa_rport_send_fwcreate(rp); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_fwc_del); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_reqq_wcancel(&rp->reqq_wait); bfa_rport_free(rp); break; case BFA_RPORT_SM_OFFLINE: bfa_stats(rp, sm_fwc_off); bfa_sm_set_state(rp, bfa_rport_sm_offline); bfa_reqq_wcancel(&rp->reqq_wait); bfa_rport_offline_cb(rp); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_fwc_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); bfa_reqq_wcancel(&rp->reqq_wait); break; default: bfa_stats(rp, sm_fwc_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Online state - normal parking state. */ static void bfa_rport_sm_online(struct bfa_rport_s *rp, enum bfa_rport_event event) { struct bfi_rport_qos_scn_s *qos_scn; bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_OFFLINE: bfa_stats(rp, sm_on_off); if (bfa_rport_send_fwdelete(rp)) bfa_sm_set_state(rp, bfa_rport_sm_fwdelete); else bfa_sm_set_state(rp, bfa_rport_sm_fwdelete_qfull); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_on_del); if (bfa_rport_send_fwdelete(rp)) bfa_sm_set_state(rp, bfa_rport_sm_deleting); else bfa_sm_set_state(rp, bfa_rport_sm_deleting_qfull); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_on_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); break; case BFA_RPORT_SM_SET_SPEED: bfa_rport_send_fwspeed(rp); break; case BFA_RPORT_SM_QOS_SCN: qos_scn = (struct bfi_rport_qos_scn_s *) rp->event_arg.fw_msg; rp->qos_attr = qos_scn->new_qos_attr; bfa_trc(rp->bfa, qos_scn->old_qos_attr.qos_flow_id); bfa_trc(rp->bfa, qos_scn->new_qos_attr.qos_flow_id); bfa_trc(rp->bfa, qos_scn->old_qos_attr.qos_priority); bfa_trc(rp->bfa, qos_scn->new_qos_attr.qos_priority); qos_scn->old_qos_attr.qos_flow_id = be32_to_cpu(qos_scn->old_qos_attr.qos_flow_id); qos_scn->new_qos_attr.qos_flow_id = be32_to_cpu(qos_scn->new_qos_attr.qos_flow_id); if (qos_scn->old_qos_attr.qos_flow_id != qos_scn->new_qos_attr.qos_flow_id) bfa_cb_rport_qos_scn_flowid(rp->rport_drv, qos_scn->old_qos_attr, qos_scn->new_qos_attr); if (qos_scn->old_qos_attr.qos_priority != qos_scn->new_qos_attr.qos_priority) bfa_cb_rport_qos_scn_prio(rp->rport_drv, qos_scn->old_qos_attr, qos_scn->new_qos_attr); break; default: bfa_stats(rp, sm_on_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Firmware rport is being deleted - awaiting f/w response. */ static void bfa_rport_sm_fwdelete(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_FWRSP: bfa_stats(rp, sm_fwd_rsp); bfa_sm_set_state(rp, bfa_rport_sm_offline); bfa_rport_offline_cb(rp); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_fwd_del); bfa_sm_set_state(rp, bfa_rport_sm_deleting); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_fwd_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); bfa_rport_offline_cb(rp); break; default: bfa_stats(rp, sm_fwd_unexp); bfa_sm_fault(rp->bfa, event); } } static void bfa_rport_sm_fwdelete_qfull(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_QRESUME: bfa_sm_set_state(rp, bfa_rport_sm_fwdelete); bfa_rport_send_fwdelete(rp); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_fwd_del); bfa_sm_set_state(rp, bfa_rport_sm_deleting_qfull); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_fwd_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); bfa_reqq_wcancel(&rp->reqq_wait); bfa_rport_offline_cb(rp); break; default: bfa_stats(rp, sm_fwd_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Offline state. */ static void bfa_rport_sm_offline(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_off_del); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_rport_free(rp); break; case BFA_RPORT_SM_ONLINE: bfa_stats(rp, sm_off_on); if (bfa_rport_send_fwcreate(rp)) bfa_sm_set_state(rp, bfa_rport_sm_fwcreate); else bfa_sm_set_state(rp, bfa_rport_sm_fwcreate_qfull); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_off_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); break; case BFA_RPORT_SM_OFFLINE: bfa_rport_offline_cb(rp); break; default: bfa_stats(rp, sm_off_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Rport is deleted, waiting for firmware response to delete. */ static void bfa_rport_sm_deleting(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_FWRSP: bfa_stats(rp, sm_del_fwrsp); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_rport_free(rp); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_del_hwf); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_rport_free(rp); break; default: bfa_sm_fault(rp->bfa, event); } } static void bfa_rport_sm_deleting_qfull(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_QRESUME: bfa_stats(rp, sm_del_fwrsp); bfa_sm_set_state(rp, bfa_rport_sm_deleting); bfa_rport_send_fwdelete(rp); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_del_hwf); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_reqq_wcancel(&rp->reqq_wait); bfa_rport_free(rp); break; default: bfa_sm_fault(rp->bfa, event); } } /* * Waiting for rport create response from firmware. A delete is pending. */ static void bfa_rport_sm_delete_pending(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_FWRSP: bfa_stats(rp, sm_delp_fwrsp); if (bfa_rport_send_fwdelete(rp)) bfa_sm_set_state(rp, bfa_rport_sm_deleting); else bfa_sm_set_state(rp, bfa_rport_sm_deleting_qfull); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_delp_hwf); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_rport_free(rp); break; default: bfa_stats(rp, sm_delp_unexp); bfa_sm_fault(rp->bfa, event); } } /* * Waiting for rport create response from firmware. Rport offline is pending. */ static void bfa_rport_sm_offline_pending(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_FWRSP: bfa_stats(rp, sm_offp_fwrsp); if (bfa_rport_send_fwdelete(rp)) bfa_sm_set_state(rp, bfa_rport_sm_fwdelete); else bfa_sm_set_state(rp, bfa_rport_sm_fwdelete_qfull); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_offp_del); bfa_sm_set_state(rp, bfa_rport_sm_delete_pending); break; case BFA_RPORT_SM_HWFAIL: bfa_stats(rp, sm_offp_hwf); bfa_sm_set_state(rp, bfa_rport_sm_iocdisable); bfa_rport_offline_cb(rp); break; default: bfa_stats(rp, sm_offp_unexp); bfa_sm_fault(rp->bfa, event); } } /* * IOC h/w failed. */ static void bfa_rport_sm_iocdisable(struct bfa_rport_s *rp, enum bfa_rport_event event) { bfa_trc(rp->bfa, rp->rport_tag); bfa_trc(rp->bfa, event); switch (event) { case BFA_RPORT_SM_OFFLINE: bfa_stats(rp, sm_iocd_off); bfa_rport_offline_cb(rp); break; case BFA_RPORT_SM_DELETE: bfa_stats(rp, sm_iocd_del); bfa_sm_set_state(rp, bfa_rport_sm_uninit); bfa_rport_free(rp); break; case BFA_RPORT_SM_ONLINE: bfa_stats(rp, sm_iocd_on); if (bfa_rport_send_fwcreate(rp)) bfa_sm_set_state(rp, bfa_rport_sm_fwcreate); else bfa_sm_set_state(rp, bfa_rport_sm_fwcreate_qfull); break; case BFA_RPORT_SM_HWFAIL: break; default: bfa_stats(rp, sm_iocd_unexp); bfa_sm_fault(rp->bfa, event); } } /* * bfa_rport_private BFA rport private functions */ static void __bfa_cb_rport_online(void *cbarg, bfa_boolean_t complete) { struct bfa_rport_s *rp = cbarg; if (complete) bfa_cb_rport_online(rp->rport_drv); } static void __bfa_cb_rport_offline(void *cbarg, bfa_boolean_t complete) { struct bfa_rport_s *rp = cbarg; if (complete) bfa_cb_rport_offline(rp->rport_drv); } static void bfa_rport_qresume(void *cbarg) { struct bfa_rport_s *rp = cbarg; bfa_sm_send_event(rp, BFA_RPORT_SM_QRESUME); } void bfa_rport_meminfo(struct bfa_iocfc_cfg_s *cfg, struct bfa_meminfo_s *minfo, struct bfa_s *bfa) { struct bfa_mem_kva_s *rport_kva = BFA_MEM_RPORT_KVA(bfa); if (cfg->fwcfg.num_rports < BFA_RPORT_MIN) cfg->fwcfg.num_rports = BFA_RPORT_MIN; /* kva memory */ bfa_mem_kva_setup(minfo, rport_kva, cfg->fwcfg.num_rports * sizeof(struct bfa_rport_s)); } void bfa_rport_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_rport_mod_s *mod = BFA_RPORT_MOD(bfa); struct bfa_rport_s *rp; u16 i; INIT_LIST_HEAD(&mod->rp_free_q); INIT_LIST_HEAD(&mod->rp_active_q); INIT_LIST_HEAD(&mod->rp_unused_q); rp = (struct bfa_rport_s *) bfa_mem_kva_curp(mod); mod->rps_list = rp; mod->num_rports = cfg->fwcfg.num_rports; WARN_ON(!mod->num_rports || (mod->num_rports & (mod->num_rports - 1))); for (i = 0; i < mod->num_rports; i++, rp++) { memset(rp, 0, sizeof(struct bfa_rport_s)); rp->bfa = bfa; rp->rport_tag = i; bfa_sm_set_state(rp, bfa_rport_sm_uninit); /* * - is unused */ if (i) list_add_tail(&rp->qe, &mod->rp_free_q); bfa_reqq_winit(&rp->reqq_wait, bfa_rport_qresume, rp); } /* * consume memory */ bfa_mem_kva_curp(mod) = (u8 *) rp; } void bfa_rport_iocdisable(struct bfa_s *bfa) { struct bfa_rport_mod_s *mod = BFA_RPORT_MOD(bfa); struct bfa_rport_s *rport; struct list_head *qe, *qen; /* Enqueue unused rport resources to free_q */ list_splice_tail_init(&mod->rp_unused_q, &mod->rp_free_q); list_for_each_safe(qe, qen, &mod->rp_active_q) { rport = (struct bfa_rport_s *) qe; bfa_sm_send_event(rport, BFA_RPORT_SM_HWFAIL); } } static struct bfa_rport_s * bfa_rport_alloc(struct bfa_rport_mod_s *mod) { struct bfa_rport_s *rport; bfa_q_deq(&mod->rp_free_q, &rport); if (rport) list_add_tail(&rport->qe, &mod->rp_active_q); return rport; } static void bfa_rport_free(struct bfa_rport_s *rport) { struct bfa_rport_mod_s *mod = BFA_RPORT_MOD(rport->bfa); WARN_ON(!bfa_q_is_on_q(&mod->rp_active_q, rport)); list_del(&rport->qe); list_add_tail(&rport->qe, &mod->rp_free_q); } static bfa_boolean_t bfa_rport_send_fwcreate(struct bfa_rport_s *rp) { struct bfi_rport_create_req_s *m; /* * check for room in queue to send request now */ m = bfa_reqq_next(rp->bfa, BFA_REQQ_RPORT); if (!m) { bfa_reqq_wait(rp->bfa, BFA_REQQ_RPORT, &rp->reqq_wait); return BFA_FALSE; } bfi_h2i_set(m->mh, BFI_MC_RPORT, BFI_RPORT_H2I_CREATE_REQ, bfa_fn_lpu(rp->bfa)); m->bfa_handle = rp->rport_tag; m->max_frmsz = cpu_to_be16(rp->rport_info.max_frmsz); m->pid = rp->rport_info.pid; m->lp_fwtag = bfa_lps_get_fwtag(rp->bfa, (u8)rp->rport_info.lp_tag); m->local_pid = rp->rport_info.local_pid; m->fc_class = rp->rport_info.fc_class; m->vf_en = rp->rport_info.vf_en; m->vf_id = rp->rport_info.vf_id; m->cisc = rp->rport_info.cisc; /* * queue I/O message to firmware */ bfa_reqq_produce(rp->bfa, BFA_REQQ_RPORT, m->mh); return BFA_TRUE; } static bfa_boolean_t bfa_rport_send_fwdelete(struct bfa_rport_s *rp) { struct bfi_rport_delete_req_s *m; /* * check for room in queue to send request now */ m = bfa_reqq_next(rp->bfa, BFA_REQQ_RPORT); if (!m) { bfa_reqq_wait(rp->bfa, BFA_REQQ_RPORT, &rp->reqq_wait); return BFA_FALSE; } bfi_h2i_set(m->mh, BFI_MC_RPORT, BFI_RPORT_H2I_DELETE_REQ, bfa_fn_lpu(rp->bfa)); m->fw_handle = rp->fw_handle; /* * queue I/O message to firmware */ bfa_reqq_produce(rp->bfa, BFA_REQQ_RPORT, m->mh); return BFA_TRUE; } static bfa_boolean_t bfa_rport_send_fwspeed(struct bfa_rport_s *rp) { struct bfa_rport_speed_req_s *m; /* * check for room in queue to send request now */ m = bfa_reqq_next(rp->bfa, BFA_REQQ_RPORT); if (!m) { bfa_trc(rp->bfa, rp->rport_info.speed); return BFA_FALSE; } bfi_h2i_set(m->mh, BFI_MC_RPORT, BFI_RPORT_H2I_SET_SPEED_REQ, bfa_fn_lpu(rp->bfa)); m->fw_handle = rp->fw_handle; m->speed = (u8)rp->rport_info.speed; /* * queue I/O message to firmware */ bfa_reqq_produce(rp->bfa, BFA_REQQ_RPORT, m->mh); return BFA_TRUE; } /* * bfa_rport_public */ /* * Rport interrupt processing. */ void bfa_rport_isr(struct bfa_s *bfa, struct bfi_msg_s *m) { union bfi_rport_i2h_msg_u msg; struct bfa_rport_s *rp; bfa_trc(bfa, m->mhdr.msg_id); msg.msg = m; switch (m->mhdr.msg_id) { case BFI_RPORT_I2H_CREATE_RSP: rp = BFA_RPORT_FROM_TAG(bfa, msg.create_rsp->bfa_handle); rp->fw_handle = msg.create_rsp->fw_handle; rp->qos_attr = msg.create_rsp->qos_attr; bfa_rport_set_lunmask(bfa, rp); WARN_ON(msg.create_rsp->status != BFA_STATUS_OK); bfa_sm_send_event(rp, BFA_RPORT_SM_FWRSP); break; case BFI_RPORT_I2H_DELETE_RSP: rp = BFA_RPORT_FROM_TAG(bfa, msg.delete_rsp->bfa_handle); WARN_ON(msg.delete_rsp->status != BFA_STATUS_OK); bfa_rport_unset_lunmask(bfa, rp); bfa_sm_send_event(rp, BFA_RPORT_SM_FWRSP); break; case BFI_RPORT_I2H_QOS_SCN: rp = BFA_RPORT_FROM_TAG(bfa, msg.qos_scn_evt->bfa_handle); rp->event_arg.fw_msg = msg.qos_scn_evt; bfa_sm_send_event(rp, BFA_RPORT_SM_QOS_SCN); break; case BFI_RPORT_I2H_LIP_SCN_ONLINE: bfa_fcport_update_loop_info(BFA_FCPORT_MOD(bfa), &msg.lip_scn->loop_info); bfa_cb_rport_scn_online(bfa); break; case BFI_RPORT_I2H_LIP_SCN_OFFLINE: bfa_cb_rport_scn_offline(bfa); break; case BFI_RPORT_I2H_NO_DEV: rp = BFA_RPORT_FROM_TAG(bfa, msg.lip_scn->bfa_handle); bfa_cb_rport_scn_no_dev(rp->rport_drv); break; default: bfa_trc(bfa, m->mhdr.msg_id); WARN_ON(1); } } void bfa_rport_res_recfg(struct bfa_s *bfa, u16 num_rport_fw) { struct bfa_rport_mod_s *mod = BFA_RPORT_MOD(bfa); struct list_head *qe; int i; for (i = 0; i < (mod->num_rports - num_rport_fw); i++) { bfa_q_deq_tail(&mod->rp_free_q, &qe); list_add_tail(qe, &mod->rp_unused_q); } } /* * bfa_rport_api */ struct bfa_rport_s * bfa_rport_create(struct bfa_s *bfa, void *rport_drv) { struct bfa_rport_s *rp; rp = bfa_rport_alloc(BFA_RPORT_MOD(bfa)); if (rp == NULL) return NULL; rp->bfa = bfa; rp->rport_drv = rport_drv; memset(&rp->stats, 0, sizeof(rp->stats)); WARN_ON(!bfa_sm_cmp_state(rp, bfa_rport_sm_uninit)); bfa_sm_send_event(rp, BFA_RPORT_SM_CREATE); return rp; } void bfa_rport_online(struct bfa_rport_s *rport, struct bfa_rport_info_s *rport_info) { WARN_ON(rport_info->max_frmsz == 0); /* * Some JBODs are seen to be not setting PDU size correctly in PLOGI * responses. Default to minimum size. */ if (rport_info->max_frmsz == 0) { bfa_trc(rport->bfa, rport->rport_tag); rport_info->max_frmsz = FC_MIN_PDUSZ; } rport->rport_info = *rport_info; bfa_sm_send_event(rport, BFA_RPORT_SM_ONLINE); } void bfa_rport_speed(struct bfa_rport_s *rport, enum bfa_port_speed speed) { WARN_ON(speed == 0); WARN_ON(speed == BFA_PORT_SPEED_AUTO); if (rport) { rport->rport_info.speed = speed; bfa_sm_send_event(rport, BFA_RPORT_SM_SET_SPEED); } } /* Set Rport LUN Mask */ void bfa_rport_set_lunmask(struct bfa_s *bfa, struct bfa_rport_s *rp) { struct bfa_lps_mod_s *lps_mod = BFA_LPS_MOD(bfa); wwn_t lp_wwn, rp_wwn; u8 lp_tag = (u8)rp->rport_info.lp_tag; rp_wwn = ((struct bfa_fcs_rport_s *)rp->rport_drv)->pwwn; lp_wwn = (BFA_LPS_FROM_TAG(lps_mod, rp->rport_info.lp_tag))->pwwn; BFA_LPS_FROM_TAG(lps_mod, rp->rport_info.lp_tag)->lun_mask = rp->lun_mask = BFA_TRUE; bfa_fcpim_lunmask_rp_update(bfa, lp_wwn, rp_wwn, rp->rport_tag, lp_tag); } /* Unset Rport LUN mask */ void bfa_rport_unset_lunmask(struct bfa_s *bfa, struct bfa_rport_s *rp) { struct bfa_lps_mod_s *lps_mod = BFA_LPS_MOD(bfa); wwn_t lp_wwn, rp_wwn; rp_wwn = ((struct bfa_fcs_rport_s *)rp->rport_drv)->pwwn; lp_wwn = (BFA_LPS_FROM_TAG(lps_mod, rp->rport_info.lp_tag))->pwwn; BFA_LPS_FROM_TAG(lps_mod, rp->rport_info.lp_tag)->lun_mask = rp->lun_mask = BFA_FALSE; bfa_fcpim_lunmask_rp_update(bfa, lp_wwn, rp_wwn, BFA_RPORT_TAG_INVALID, BFA_LP_TAG_INVALID); } /* * SGPG related functions */ /* * Compute and return memory needed by FCP(im) module. */ void bfa_sgpg_meminfo(struct bfa_iocfc_cfg_s *cfg, struct bfa_meminfo_s *minfo, struct bfa_s *bfa) { struct bfa_sgpg_mod_s *sgpg_mod = BFA_SGPG_MOD(bfa); struct bfa_mem_kva_s *sgpg_kva = BFA_MEM_SGPG_KVA(bfa); struct bfa_mem_dma_s *seg_ptr; u16 nsegs, idx, per_seg_sgpg, num_sgpg; u32 sgpg_sz = sizeof(struct bfi_sgpg_s); if (cfg->drvcfg.num_sgpgs < BFA_SGPG_MIN) cfg->drvcfg.num_sgpgs = BFA_SGPG_MIN; else if (cfg->drvcfg.num_sgpgs > BFA_SGPG_MAX) cfg->drvcfg.num_sgpgs = BFA_SGPG_MAX; num_sgpg = cfg->drvcfg.num_sgpgs; nsegs = BFI_MEM_DMA_NSEGS(num_sgpg, sgpg_sz); per_seg_sgpg = BFI_MEM_NREQS_SEG(sgpg_sz); bfa_mem_dma_seg_iter(sgpg_mod, seg_ptr, nsegs, idx) { if (num_sgpg >= per_seg_sgpg) { num_sgpg -= per_seg_sgpg; bfa_mem_dma_setup(minfo, seg_ptr, per_seg_sgpg * sgpg_sz); } else bfa_mem_dma_setup(minfo, seg_ptr, num_sgpg * sgpg_sz); } /* kva memory */ bfa_mem_kva_setup(minfo, sgpg_kva, cfg->drvcfg.num_sgpgs * sizeof(struct bfa_sgpg_s)); } void bfa_sgpg_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_sgpg_mod_s *mod = BFA_SGPG_MOD(bfa); struct bfa_sgpg_s *hsgpg; struct bfi_sgpg_s *sgpg; u64 align_len; struct bfa_mem_dma_s *seg_ptr; u32 sgpg_sz = sizeof(struct bfi_sgpg_s); u16 i, idx, nsegs, per_seg_sgpg, num_sgpg; union { u64 pa; union bfi_addr_u addr; } sgpg_pa, sgpg_pa_tmp; INIT_LIST_HEAD(&mod->sgpg_q); INIT_LIST_HEAD(&mod->sgpg_wait_q); bfa_trc(bfa, cfg->drvcfg.num_sgpgs); mod->free_sgpgs = mod->num_sgpgs = cfg->drvcfg.num_sgpgs; num_sgpg = cfg->drvcfg.num_sgpgs; nsegs = BFI_MEM_DMA_NSEGS(num_sgpg, sgpg_sz); /* dma/kva mem claim */ hsgpg = (struct bfa_sgpg_s *) bfa_mem_kva_curp(mod); bfa_mem_dma_seg_iter(mod, seg_ptr, nsegs, idx) { if (!bfa_mem_dma_virt(seg_ptr)) break; align_len = BFA_SGPG_ROUNDUP(bfa_mem_dma_phys(seg_ptr)) - bfa_mem_dma_phys(seg_ptr); sgpg = (struct bfi_sgpg_s *) (((u8 *) bfa_mem_dma_virt(seg_ptr)) + align_len); sgpg_pa.pa = bfa_mem_dma_phys(seg_ptr) + align_len; WARN_ON(sgpg_pa.pa & (sgpg_sz - 1)); per_seg_sgpg = (seg_ptr->mem_len - (u32)align_len) / sgpg_sz; for (i = 0; num_sgpg > 0 && i < per_seg_sgpg; i++, num_sgpg--) { memset(hsgpg, 0, sizeof(*hsgpg)); memset(sgpg, 0, sizeof(*sgpg)); hsgpg->sgpg = sgpg; sgpg_pa_tmp.pa = bfa_sgaddr_le(sgpg_pa.pa); hsgpg->sgpg_pa = sgpg_pa_tmp.addr; list_add_tail(&hsgpg->qe, &mod->sgpg_q); sgpg++; hsgpg++; sgpg_pa.pa += sgpg_sz; } } bfa_mem_kva_curp(mod) = (u8 *) hsgpg; } bfa_status_t bfa_sgpg_malloc(struct bfa_s *bfa, struct list_head *sgpg_q, int nsgpgs) { struct bfa_sgpg_mod_s *mod = BFA_SGPG_MOD(bfa); struct bfa_sgpg_s *hsgpg; int i; if (mod->free_sgpgs < nsgpgs) return BFA_STATUS_ENOMEM; for (i = 0; i < nsgpgs; i++) { bfa_q_deq(&mod->sgpg_q, &hsgpg); WARN_ON(!hsgpg); list_add_tail(&hsgpg->qe, sgpg_q); } mod->free_sgpgs -= nsgpgs; return BFA_STATUS_OK; } void bfa_sgpg_mfree(struct bfa_s *bfa, struct list_head *sgpg_q, int nsgpg) { struct bfa_sgpg_mod_s *mod = BFA_SGPG_MOD(bfa); struct bfa_sgpg_wqe_s *wqe; mod->free_sgpgs += nsgpg; WARN_ON(mod->free_sgpgs > mod->num_sgpgs); list_splice_tail_init(sgpg_q, &mod->sgpg_q); if (list_empty(&mod->sgpg_wait_q)) return; /* * satisfy as many waiting requests as possible */ do { wqe = bfa_q_first(&mod->sgpg_wait_q); if (mod->free_sgpgs < wqe->nsgpg) nsgpg = mod->free_sgpgs; else nsgpg = wqe->nsgpg; bfa_sgpg_malloc(bfa, &wqe->sgpg_q, nsgpg); wqe->nsgpg -= nsgpg; if (wqe->nsgpg == 0) { list_del(&wqe->qe); wqe->cbfn(wqe->cbarg); } } while (mod->free_sgpgs && !list_empty(&mod->sgpg_wait_q)); } void bfa_sgpg_wait(struct bfa_s *bfa, struct bfa_sgpg_wqe_s *wqe, int nsgpg) { struct bfa_sgpg_mod_s *mod = BFA_SGPG_MOD(bfa); WARN_ON(nsgpg <= 0); WARN_ON(nsgpg <= mod->free_sgpgs); wqe->nsgpg_total = wqe->nsgpg = nsgpg; /* * allocate any left to this one first */ if (mod->free_sgpgs) { /* * no one else is waiting for SGPG */ WARN_ON(!list_empty(&mod->sgpg_wait_q)); list_splice_tail_init(&mod->sgpg_q, &wqe->sgpg_q); wqe->nsgpg -= mod->free_sgpgs; mod->free_sgpgs = 0; } list_add_tail(&wqe->qe, &mod->sgpg_wait_q); } void bfa_sgpg_wcancel(struct bfa_s *bfa, struct bfa_sgpg_wqe_s *wqe) { struct bfa_sgpg_mod_s *mod = BFA_SGPG_MOD(bfa); WARN_ON(!bfa_q_is_on_q(&mod->sgpg_wait_q, wqe)); list_del(&wqe->qe); if (wqe->nsgpg_total != wqe->nsgpg) bfa_sgpg_mfree(bfa, &wqe->sgpg_q, wqe->nsgpg_total - wqe->nsgpg); } void bfa_sgpg_winit(struct bfa_sgpg_wqe_s *wqe, void (*cbfn) (void *cbarg), void *cbarg) { INIT_LIST_HEAD(&wqe->sgpg_q); wqe->cbfn = cbfn; wqe->cbarg = cbarg; } /* * UF related functions */ /* ***************************************************************************** * Internal functions ***************************************************************************** */ static void __bfa_cb_uf_recv(void *cbarg, bfa_boolean_t complete) { struct bfa_uf_s *uf = cbarg; struct bfa_uf_mod_s *ufm = BFA_UF_MOD(uf->bfa); if (complete) ufm->ufrecv(ufm->cbarg, uf); } static void claim_uf_post_msgs(struct bfa_uf_mod_s *ufm) { struct bfi_uf_buf_post_s *uf_bp_msg; u16 i; u16 buf_len; ufm->uf_buf_posts = (struct bfi_uf_buf_post_s *) bfa_mem_kva_curp(ufm); uf_bp_msg = ufm->uf_buf_posts; for (i = 0, uf_bp_msg = ufm->uf_buf_posts; i < ufm->num_ufs; i++, uf_bp_msg++) { memset(uf_bp_msg, 0, sizeof(struct bfi_uf_buf_post_s)); uf_bp_msg->buf_tag = i; buf_len = sizeof(struct bfa_uf_buf_s); uf_bp_msg->buf_len = cpu_to_be16(buf_len); bfi_h2i_set(uf_bp_msg->mh, BFI_MC_UF, BFI_UF_H2I_BUF_POST, bfa_fn_lpu(ufm->bfa)); bfa_alen_set(&uf_bp_msg->alen, buf_len, ufm_pbs_pa(ufm, i)); } /* * advance pointer beyond consumed memory */ bfa_mem_kva_curp(ufm) = (u8 *) uf_bp_msg; } static void claim_ufs(struct bfa_uf_mod_s *ufm) { u16 i; struct bfa_uf_s *uf; /* * Claim block of memory for UF list */ ufm->uf_list = (struct bfa_uf_s *) bfa_mem_kva_curp(ufm); /* * Initialize UFs and queue it in UF free queue */ for (i = 0, uf = ufm->uf_list; i < ufm->num_ufs; i++, uf++) { memset(uf, 0, sizeof(struct bfa_uf_s)); uf->bfa = ufm->bfa; uf->uf_tag = i; uf->pb_len = BFA_PER_UF_DMA_SZ; uf->buf_kva = bfa_mem_get_dmabuf_kva(ufm, i, BFA_PER_UF_DMA_SZ); uf->buf_pa = ufm_pbs_pa(ufm, i); list_add_tail(&uf->qe, &ufm->uf_free_q); } /* * advance memory pointer */ bfa_mem_kva_curp(ufm) = (u8 *) uf; } static void uf_mem_claim(struct bfa_uf_mod_s *ufm) { claim_ufs(ufm); claim_uf_post_msgs(ufm); } void bfa_uf_meminfo(struct bfa_iocfc_cfg_s *cfg, struct bfa_meminfo_s *minfo, struct bfa_s *bfa) { struct bfa_uf_mod_s *ufm = BFA_UF_MOD(bfa); struct bfa_mem_kva_s *uf_kva = BFA_MEM_UF_KVA(bfa); u32 num_ufs = cfg->fwcfg.num_uf_bufs; struct bfa_mem_dma_s *seg_ptr; u16 nsegs, idx, per_seg_uf = 0; nsegs = BFI_MEM_DMA_NSEGS(num_ufs, BFA_PER_UF_DMA_SZ); per_seg_uf = BFI_MEM_NREQS_SEG(BFA_PER_UF_DMA_SZ); bfa_mem_dma_seg_iter(ufm, seg_ptr, nsegs, idx) { if (num_ufs >= per_seg_uf) { num_ufs -= per_seg_uf; bfa_mem_dma_setup(minfo, seg_ptr, per_seg_uf * BFA_PER_UF_DMA_SZ); } else bfa_mem_dma_setup(minfo, seg_ptr, num_ufs * BFA_PER_UF_DMA_SZ); } /* kva memory */ bfa_mem_kva_setup(minfo, uf_kva, cfg->fwcfg.num_uf_bufs * (sizeof(struct bfa_uf_s) + sizeof(struct bfi_uf_buf_post_s))); } void bfa_uf_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_uf_mod_s *ufm = BFA_UF_MOD(bfa); ufm->bfa = bfa; ufm->num_ufs = cfg->fwcfg.num_uf_bufs; INIT_LIST_HEAD(&ufm->uf_free_q); INIT_LIST_HEAD(&ufm->uf_posted_q); INIT_LIST_HEAD(&ufm->uf_unused_q); uf_mem_claim(ufm); } static struct bfa_uf_s * bfa_uf_get(struct bfa_uf_mod_s *uf_mod) { struct bfa_uf_s *uf; bfa_q_deq(&uf_mod->uf_free_q, &uf); return uf; } static void bfa_uf_put(struct bfa_uf_mod_s *uf_mod, struct bfa_uf_s *uf) { list_add_tail(&uf->qe, &uf_mod->uf_free_q); } static bfa_status_t bfa_uf_post(struct bfa_uf_mod_s *ufm, struct bfa_uf_s *uf) { struct bfi_uf_buf_post_s *uf_post_msg; uf_post_msg = bfa_reqq_next(ufm->bfa, BFA_REQQ_FCXP); if (!uf_post_msg) return BFA_STATUS_FAILED; memcpy(uf_post_msg, &ufm->uf_buf_posts[uf->uf_tag], sizeof(struct bfi_uf_buf_post_s)); bfa_reqq_produce(ufm->bfa, BFA_REQQ_FCXP, uf_post_msg->mh); bfa_trc(ufm->bfa, uf->uf_tag); list_add_tail(&uf->qe, &ufm->uf_posted_q); return BFA_STATUS_OK; } static void bfa_uf_post_all(struct bfa_uf_mod_s *uf_mod) { struct bfa_uf_s *uf; while ((uf = bfa_uf_get(uf_mod)) != NULL) { if (bfa_uf_post(uf_mod, uf) != BFA_STATUS_OK) break; } } static void uf_recv(struct bfa_s *bfa, struct bfi_uf_frm_rcvd_s *m) { struct bfa_uf_mod_s *ufm = BFA_UF_MOD(bfa); u16 uf_tag = m->buf_tag; struct bfa_uf_s *uf = &ufm->uf_list[uf_tag]; struct bfa_uf_buf_s *uf_buf; uint8_t *buf; struct fchs_s *fchs; uf_buf = (struct bfa_uf_buf_s *) bfa_mem_get_dmabuf_kva(ufm, uf_tag, uf->pb_len); buf = &uf_buf->d[0]; m->frm_len = be16_to_cpu(m->frm_len); m->xfr_len = be16_to_cpu(m->xfr_len); fchs = (struct fchs_s *)uf_buf; list_del(&uf->qe); /* dequeue from posted queue */ uf->data_ptr = buf; uf->data_len = m->xfr_len; WARN_ON(uf->data_len < sizeof(struct fchs_s)); if (uf->data_len == sizeof(struct fchs_s)) { bfa_plog_fchdr(bfa->plog, BFA_PL_MID_HAL_UF, BFA_PL_EID_RX, uf->data_len, (struct fchs_s *)buf); } else { u32 pld_w0 = *((u32 *) (buf + sizeof(struct fchs_s))); bfa_plog_fchdr_and_pl(bfa->plog, BFA_PL_MID_HAL_UF, BFA_PL_EID_RX, uf->data_len, (struct fchs_s *)buf, pld_w0); } if (bfa->fcs) __bfa_cb_uf_recv(uf, BFA_TRUE); else bfa_cb_queue(bfa, &uf->hcb_qe, __bfa_cb_uf_recv, uf); } void bfa_uf_iocdisable(struct bfa_s *bfa) { struct bfa_uf_mod_s *ufm = BFA_UF_MOD(bfa); struct bfa_uf_s *uf; struct list_head *qe, *qen; /* Enqueue unused uf resources to free_q */ list_splice_tail_init(&ufm->uf_unused_q, &ufm->uf_free_q); list_for_each_safe(qe, qen, &ufm->uf_posted_q) { uf = (struct bfa_uf_s *) qe; list_del(&uf->qe); bfa_uf_put(ufm, uf); } } void bfa_uf_start(struct bfa_s *bfa) { bfa_uf_post_all(BFA_UF_MOD(bfa)); } /* * Register handler for all unsolicted receive frames. * * @param[in] bfa BFA instance * @param[in] ufrecv receive handler function * @param[in] cbarg receive handler arg */ void bfa_uf_recv_register(struct bfa_s *bfa, bfa_cb_uf_recv_t ufrecv, void *cbarg) { struct bfa_uf_mod_s *ufm = BFA_UF_MOD(bfa); ufm->ufrecv = ufrecv; ufm->cbarg = cbarg; } /* * Free an unsolicited frame back to BFA. * * @param[in] uf unsolicited frame to be freed * * @return None */ void bfa_uf_free(struct bfa_uf_s *uf) { bfa_uf_put(BFA_UF_MOD(uf->bfa), uf); bfa_uf_post_all(BFA_UF_MOD(uf->bfa)); } /* * uf_pub BFA uf module public functions */ void bfa_uf_isr(struct bfa_s *bfa, struct bfi_msg_s *msg) { bfa_trc(bfa, msg->mhdr.msg_id); switch (msg->mhdr.msg_id) { case BFI_UF_I2H_FRM_RCVD: uf_recv(bfa, (struct bfi_uf_frm_rcvd_s *) msg); break; default: bfa_trc(bfa, msg->mhdr.msg_id); WARN_ON(1); } } void bfa_uf_res_recfg(struct bfa_s *bfa, u16 num_uf_fw) { struct bfa_uf_mod_s *mod = BFA_UF_MOD(bfa); struct list_head *qe; int i; for (i = 0; i < (mod->num_ufs - num_uf_fw); i++) { bfa_q_deq_tail(&mod->uf_free_q, &qe); list_add_tail(qe, &mod->uf_unused_q); } } /* * Dport forward declaration */ enum bfa_dport_test_state_e { BFA_DPORT_ST_DISABLED = 0, /*!< dport is disabled */ BFA_DPORT_ST_INP = 1, /*!< test in progress */ BFA_DPORT_ST_COMP = 2, /*!< test complete successfully */ BFA_DPORT_ST_NO_SFP = 3, /*!< sfp is not present */ BFA_DPORT_ST_NOTSTART = 4, /*!< test not start dport is enabled */ }; /* * BFA DPORT state machine events */ enum bfa_dport_sm_event { BFA_DPORT_SM_ENABLE = 1, /* dport enable event */ BFA_DPORT_SM_DISABLE = 2, /* dport disable event */ BFA_DPORT_SM_FWRSP = 3, /* fw enable/disable rsp */ BFA_DPORT_SM_QRESUME = 4, /* CQ space available */ BFA_DPORT_SM_HWFAIL = 5, /* IOC h/w failure */ BFA_DPORT_SM_START = 6, /* re-start dport test */ BFA_DPORT_SM_REQFAIL = 7, /* request failure */ BFA_DPORT_SM_SCN = 8, /* state change notify frm fw */ }; static void bfa_dport_sm_disabled(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_enabling_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_enabling(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_enabled(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_disabling_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_disabling(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_starting_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_starting(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_dynamic_disabling(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_sm_dynamic_disabling_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event); static void bfa_dport_qresume(void *cbarg); static void bfa_dport_req_comp(struct bfa_dport_s *dport, struct bfi_diag_dport_rsp_s *msg); static void bfa_dport_scn(struct bfa_dport_s *dport, struct bfi_diag_dport_scn_s *msg); /* * BFA fcdiag module */ #define BFA_DIAG_QTEST_TOV 1000 /* msec */ /* * Set port status to busy */ static void bfa_fcdiag_set_busy_status(struct bfa_fcdiag_s *fcdiag) { struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(fcdiag->bfa); if (fcdiag->lb.lock) fcport->diag_busy = BFA_TRUE; else fcport->diag_busy = BFA_FALSE; } void bfa_fcdiag_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_pcidev_s *pcidev) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); struct bfa_dport_s *dport = &fcdiag->dport; fcdiag->bfa = bfa; fcdiag->trcmod = bfa->trcmod; /* The common DIAG attach bfa_diag_attach() will do all memory claim */ dport->bfa = bfa; bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_reqq_winit(&dport->reqq_wait, bfa_dport_qresume, dport); dport->cbfn = NULL; dport->cbarg = NULL; dport->test_state = BFA_DPORT_ST_DISABLED; memset(&dport->result, 0, sizeof(struct bfa_diag_dport_result_s)); } void bfa_fcdiag_iocdisable(struct bfa_s *bfa) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); struct bfa_dport_s *dport = &fcdiag->dport; bfa_trc(fcdiag, fcdiag->lb.lock); if (fcdiag->lb.lock) { fcdiag->lb.status = BFA_STATUS_IOC_FAILURE; fcdiag->lb.cbfn(fcdiag->lb.cbarg, fcdiag->lb.status); fcdiag->lb.lock = 0; bfa_fcdiag_set_busy_status(fcdiag); } bfa_sm_send_event(dport, BFA_DPORT_SM_HWFAIL); } static void bfa_fcdiag_queuetest_timeout(void *cbarg) { struct bfa_fcdiag_s *fcdiag = cbarg; struct bfa_diag_qtest_result_s *res = fcdiag->qtest.result; bfa_trc(fcdiag, fcdiag->qtest.all); bfa_trc(fcdiag, fcdiag->qtest.count); fcdiag->qtest.timer_active = 0; res->status = BFA_STATUS_ETIMER; res->count = QTEST_CNT_DEFAULT - fcdiag->qtest.count; if (fcdiag->qtest.all) res->queue = fcdiag->qtest.all; bfa_trc(fcdiag, BFA_STATUS_ETIMER); fcdiag->qtest.status = BFA_STATUS_ETIMER; fcdiag->qtest.cbfn(fcdiag->qtest.cbarg, fcdiag->qtest.status); fcdiag->qtest.lock = 0; } static bfa_status_t bfa_fcdiag_queuetest_send(struct bfa_fcdiag_s *fcdiag) { u32 i; struct bfi_diag_qtest_req_s *req; req = bfa_reqq_next(fcdiag->bfa, fcdiag->qtest.queue); if (!req) return BFA_STATUS_DEVBUSY; /* build host command */ bfi_h2i_set(req->mh, BFI_MC_DIAG, BFI_DIAG_H2I_QTEST, bfa_fn_lpu(fcdiag->bfa)); for (i = 0; i < BFI_LMSG_PL_WSZ; i++) req->data[i] = QTEST_PAT_DEFAULT; bfa_trc(fcdiag, fcdiag->qtest.queue); /* ring door bell */ bfa_reqq_produce(fcdiag->bfa, fcdiag->qtest.queue, req->mh); return BFA_STATUS_OK; } static void bfa_fcdiag_queuetest_comp(struct bfa_fcdiag_s *fcdiag, bfi_diag_qtest_rsp_t *rsp) { struct bfa_diag_qtest_result_s *res = fcdiag->qtest.result; bfa_status_t status = BFA_STATUS_OK; int i; /* Check timer, should still be active */ if (!fcdiag->qtest.timer_active) { bfa_trc(fcdiag, fcdiag->qtest.timer_active); return; } /* update count */ fcdiag->qtest.count--; /* Check result */ for (i = 0; i < BFI_LMSG_PL_WSZ; i++) { if (rsp->data[i] != ~(QTEST_PAT_DEFAULT)) { res->status = BFA_STATUS_DATACORRUPTED; break; } } if (res->status == BFA_STATUS_OK) { if (fcdiag->qtest.count > 0) { status = bfa_fcdiag_queuetest_send(fcdiag); if (status == BFA_STATUS_OK) return; else res->status = status; } else if (fcdiag->qtest.all > 0 && fcdiag->qtest.queue < (BFI_IOC_MAX_CQS - 1)) { fcdiag->qtest.count = QTEST_CNT_DEFAULT; fcdiag->qtest.queue++; status = bfa_fcdiag_queuetest_send(fcdiag); if (status == BFA_STATUS_OK) return; else res->status = status; } } /* Stop timer when we comp all queue */ if (fcdiag->qtest.timer_active) { bfa_timer_stop(&fcdiag->qtest.timer); fcdiag->qtest.timer_active = 0; } res->queue = fcdiag->qtest.queue; res->count = QTEST_CNT_DEFAULT - fcdiag->qtest.count; bfa_trc(fcdiag, res->count); bfa_trc(fcdiag, res->status); fcdiag->qtest.status = res->status; fcdiag->qtest.cbfn(fcdiag->qtest.cbarg, fcdiag->qtest.status); fcdiag->qtest.lock = 0; } static void bfa_fcdiag_loopback_comp(struct bfa_fcdiag_s *fcdiag, struct bfi_diag_lb_rsp_s *rsp) { struct bfa_diag_loopback_result_s *res = fcdiag->lb.result; res->numtxmfrm = be32_to_cpu(rsp->res.numtxmfrm); res->numosffrm = be32_to_cpu(rsp->res.numosffrm); res->numrcvfrm = be32_to_cpu(rsp->res.numrcvfrm); res->badfrminf = be32_to_cpu(rsp->res.badfrminf); res->badfrmnum = be32_to_cpu(rsp->res.badfrmnum); res->status = rsp->res.status; fcdiag->lb.status = rsp->res.status; bfa_trc(fcdiag, fcdiag->lb.status); fcdiag->lb.cbfn(fcdiag->lb.cbarg, fcdiag->lb.status); fcdiag->lb.lock = 0; bfa_fcdiag_set_busy_status(fcdiag); } static bfa_status_t bfa_fcdiag_loopback_send(struct bfa_fcdiag_s *fcdiag, struct bfa_diag_loopback_s *loopback) { struct bfi_diag_lb_req_s *lb_req; lb_req = bfa_reqq_next(fcdiag->bfa, BFA_REQQ_DIAG); if (!lb_req) return BFA_STATUS_DEVBUSY; /* build host command */ bfi_h2i_set(lb_req->mh, BFI_MC_DIAG, BFI_DIAG_H2I_LOOPBACK, bfa_fn_lpu(fcdiag->bfa)); lb_req->lb_mode = loopback->lb_mode; lb_req->speed = loopback->speed; lb_req->loopcnt = loopback->loopcnt; lb_req->pattern = loopback->pattern; /* ring door bell */ bfa_reqq_produce(fcdiag->bfa, BFA_REQQ_DIAG, lb_req->mh); bfa_trc(fcdiag, loopback->lb_mode); bfa_trc(fcdiag, loopback->speed); bfa_trc(fcdiag, loopback->loopcnt); bfa_trc(fcdiag, loopback->pattern); return BFA_STATUS_OK; } /* * cpe/rme intr handler */ void bfa_fcdiag_intr(struct bfa_s *bfa, struct bfi_msg_s *msg) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); switch (msg->mhdr.msg_id) { case BFI_DIAG_I2H_LOOPBACK: bfa_fcdiag_loopback_comp(fcdiag, (struct bfi_diag_lb_rsp_s *) msg); break; case BFI_DIAG_I2H_QTEST: bfa_fcdiag_queuetest_comp(fcdiag, (bfi_diag_qtest_rsp_t *)msg); break; case BFI_DIAG_I2H_DPORT: bfa_dport_req_comp(&fcdiag->dport, (struct bfi_diag_dport_rsp_s *)msg); break; case BFI_DIAG_I2H_DPORT_SCN: bfa_dport_scn(&fcdiag->dport, (struct bfi_diag_dport_scn_s *)msg); break; default: bfa_trc(fcdiag, msg->mhdr.msg_id); WARN_ON(1); } } /* * Loopback test * * @param[in] *bfa - bfa data struct * @param[in] opmode - port operation mode * @param[in] speed - port speed * @param[in] lpcnt - loop count * @param[in] pat - pattern to build packet * @param[in] *result - pt to bfa_diag_loopback_result_t data struct * @param[in] cbfn - callback function * @param[in] cbarg - callback functioin arg * * @param[out] */ bfa_status_t bfa_fcdiag_loopback(struct bfa_s *bfa, enum bfa_port_opmode opmode, enum bfa_port_speed speed, u32 lpcnt, u32 pat, struct bfa_diag_loopback_result_s *result, bfa_cb_diag_t cbfn, void *cbarg) { struct bfa_diag_loopback_s loopback; struct bfa_port_attr_s attr; bfa_status_t status; struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); if (!bfa_iocfc_is_operational(bfa)) return BFA_STATUS_IOC_NON_OP; /* if port is PBC disabled, return error */ if (bfa_fcport_is_pbcdisabled(bfa)) { bfa_trc(fcdiag, BFA_STATUS_PBC); return BFA_STATUS_PBC; } if (bfa_fcport_is_disabled(bfa) == BFA_FALSE) { bfa_trc(fcdiag, opmode); return BFA_STATUS_PORT_NOT_DISABLED; } /* * Check if input speed is supported by the port mode */ if (bfa_ioc_get_type(&bfa->ioc) == BFA_IOC_TYPE_FC) { if (!(speed == BFA_PORT_SPEED_1GBPS || speed == BFA_PORT_SPEED_2GBPS || speed == BFA_PORT_SPEED_4GBPS || speed == BFA_PORT_SPEED_8GBPS || speed == BFA_PORT_SPEED_16GBPS || speed == BFA_PORT_SPEED_AUTO)) { bfa_trc(fcdiag, speed); return BFA_STATUS_UNSUPP_SPEED; } bfa_fcport_get_attr(bfa, &attr); bfa_trc(fcdiag, attr.speed_supported); if (speed > attr.speed_supported) return BFA_STATUS_UNSUPP_SPEED; } else { if (speed != BFA_PORT_SPEED_10GBPS) { bfa_trc(fcdiag, speed); return BFA_STATUS_UNSUPP_SPEED; } } /* * For CT2, 1G is not supported */ if ((speed == BFA_PORT_SPEED_1GBPS) && (bfa_asic_id_ct2(bfa->ioc.pcidev.device_id))) { bfa_trc(fcdiag, speed); return BFA_STATUS_UNSUPP_SPEED; } /* For Mezz card, port speed entered needs to be checked */ if (bfa_mfg_is_mezz(bfa->ioc.attr->card_type)) { if (bfa_ioc_get_type(&bfa->ioc) == BFA_IOC_TYPE_FC) { if (!(speed == BFA_PORT_SPEED_1GBPS || speed == BFA_PORT_SPEED_2GBPS || speed == BFA_PORT_SPEED_4GBPS || speed == BFA_PORT_SPEED_8GBPS || speed == BFA_PORT_SPEED_16GBPS || speed == BFA_PORT_SPEED_AUTO)) return BFA_STATUS_UNSUPP_SPEED; } else { if (speed != BFA_PORT_SPEED_10GBPS) return BFA_STATUS_UNSUPP_SPEED; } } /* check to see if fcport is dport */ if (bfa_fcport_is_dport(bfa)) { bfa_trc(fcdiag, fcdiag->lb.lock); return BFA_STATUS_DPORT_ENABLED; } /* check to see if there is another destructive diag cmd running */ if (fcdiag->lb.lock) { bfa_trc(fcdiag, fcdiag->lb.lock); return BFA_STATUS_DEVBUSY; } fcdiag->lb.lock = 1; loopback.lb_mode = opmode; loopback.speed = speed; loopback.loopcnt = lpcnt; loopback.pattern = pat; fcdiag->lb.result = result; fcdiag->lb.cbfn = cbfn; fcdiag->lb.cbarg = cbarg; memset(result, 0, sizeof(struct bfa_diag_loopback_result_s)); bfa_fcdiag_set_busy_status(fcdiag); /* Send msg to fw */ status = bfa_fcdiag_loopback_send(fcdiag, &loopback); return status; } /* * DIAG queue test command * * @param[in] *bfa - bfa data struct * @param[in] force - 1: don't do ioc op checking * @param[in] queue - queue no. to test * @param[in] *result - pt to bfa_diag_qtest_result_t data struct * @param[in] cbfn - callback function * @param[in] *cbarg - callback functioin arg * * @param[out] */ bfa_status_t bfa_fcdiag_queuetest(struct bfa_s *bfa, u32 force, u32 queue, struct bfa_diag_qtest_result_s *result, bfa_cb_diag_t cbfn, void *cbarg) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); bfa_status_t status; bfa_trc(fcdiag, force); bfa_trc(fcdiag, queue); if (!force && !bfa_iocfc_is_operational(bfa)) return BFA_STATUS_IOC_NON_OP; /* check to see if there is another destructive diag cmd running */ if (fcdiag->qtest.lock) { bfa_trc(fcdiag, fcdiag->qtest.lock); return BFA_STATUS_DEVBUSY; } /* Initialization */ fcdiag->qtest.lock = 1; fcdiag->qtest.cbfn = cbfn; fcdiag->qtest.cbarg = cbarg; fcdiag->qtest.result = result; fcdiag->qtest.count = QTEST_CNT_DEFAULT; /* Init test results */ fcdiag->qtest.result->status = BFA_STATUS_OK; fcdiag->qtest.result->count = 0; /* send */ if (queue < BFI_IOC_MAX_CQS) { fcdiag->qtest.result->queue = (u8)queue; fcdiag->qtest.queue = (u8)queue; fcdiag->qtest.all = 0; } else { fcdiag->qtest.result->queue = 0; fcdiag->qtest.queue = 0; fcdiag->qtest.all = 1; } status = bfa_fcdiag_queuetest_send(fcdiag); /* Start a timer */ if (status == BFA_STATUS_OK) { bfa_timer_start(bfa, &fcdiag->qtest.timer, bfa_fcdiag_queuetest_timeout, fcdiag, BFA_DIAG_QTEST_TOV); fcdiag->qtest.timer_active = 1; } return status; } /* * DIAG PLB is running * * @param[in] *bfa - bfa data struct * * @param[out] */ bfa_status_t bfa_fcdiag_lb_is_running(struct bfa_s *bfa) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); return fcdiag->lb.lock ? BFA_STATUS_DIAG_BUSY : BFA_STATUS_OK; } /* * D-port */ #define bfa_dport_result_start(__dport, __mode) do { \ (__dport)->result.start_time = bfa_get_log_time(); \ (__dport)->result.status = DPORT_TEST_ST_INPRG; \ (__dport)->result.mode = (__mode); \ (__dport)->result.rp_pwwn = (__dport)->rp_pwwn; \ (__dport)->result.rp_nwwn = (__dport)->rp_nwwn; \ (__dport)->result.lpcnt = (__dport)->lpcnt; \ } while (0) static bfa_boolean_t bfa_dport_send_req(struct bfa_dport_s *dport, enum bfi_dport_req req); static void bfa_cb_fcdiag_dport(struct bfa_dport_s *dport, bfa_status_t bfa_status) { if (dport->cbfn != NULL) { dport->cbfn(dport->cbarg, bfa_status); dport->cbfn = NULL; dport->cbarg = NULL; } } static void bfa_dport_sm_disabled(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_ENABLE: bfa_fcport_dportenable(dport->bfa); if (bfa_dport_send_req(dport, BFI_DPORT_ENABLE)) bfa_sm_set_state(dport, bfa_dport_sm_enabling); else bfa_sm_set_state(dport, bfa_dport_sm_enabling_qwait); break; case BFA_DPORT_SM_DISABLE: /* Already disabled */ break; case BFA_DPORT_SM_HWFAIL: /* ignore */ break; case BFA_DPORT_SM_SCN: if (dport->i2hmsg.scn.state == BFI_DPORT_SCN_DDPORT_ENABLE) { bfa_fcport_ddportenable(dport->bfa); dport->dynamic = BFA_TRUE; dport->test_state = BFA_DPORT_ST_NOTSTART; bfa_sm_set_state(dport, bfa_dport_sm_enabled); } else { bfa_trc(dport->bfa, dport->i2hmsg.scn.state); WARN_ON(1); } break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_enabling_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_QRESUME: bfa_sm_set_state(dport, bfa_dport_sm_enabling); bfa_dport_send_req(dport, BFI_DPORT_ENABLE); break; case BFA_DPORT_SM_HWFAIL: bfa_reqq_wcancel(&dport->reqq_wait); bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_cb_fcdiag_dport(dport, BFA_STATUS_FAILED); break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_enabling(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_FWRSP: memset(&dport->result, 0, sizeof(struct bfa_diag_dport_result_s)); if (dport->i2hmsg.rsp.status == BFA_STATUS_DPORT_INV_SFP) { dport->test_state = BFA_DPORT_ST_NO_SFP; } else { dport->test_state = BFA_DPORT_ST_INP; bfa_dport_result_start(dport, BFA_DPORT_OPMODE_AUTO); } bfa_sm_set_state(dport, bfa_dport_sm_enabled); break; case BFA_DPORT_SM_REQFAIL: dport->test_state = BFA_DPORT_ST_DISABLED; bfa_fcport_dportdisable(dport->bfa); bfa_sm_set_state(dport, bfa_dport_sm_disabled); break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_cb_fcdiag_dport(dport, BFA_STATUS_FAILED); break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_enabled(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_START: if (bfa_dport_send_req(dport, BFI_DPORT_START)) bfa_sm_set_state(dport, bfa_dport_sm_starting); else bfa_sm_set_state(dport, bfa_dport_sm_starting_qwait); break; case BFA_DPORT_SM_DISABLE: bfa_fcport_dportdisable(dport->bfa); if (bfa_dport_send_req(dport, BFI_DPORT_DISABLE)) bfa_sm_set_state(dport, bfa_dport_sm_disabling); else bfa_sm_set_state(dport, bfa_dport_sm_disabling_qwait); break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); break; case BFA_DPORT_SM_SCN: switch (dport->i2hmsg.scn.state) { case BFI_DPORT_SCN_TESTCOMP: dport->test_state = BFA_DPORT_ST_COMP; break; case BFI_DPORT_SCN_TESTSTART: dport->test_state = BFA_DPORT_ST_INP; break; case BFI_DPORT_SCN_TESTSKIP: case BFI_DPORT_SCN_SUBTESTSTART: /* no state change */ break; case BFI_DPORT_SCN_SFP_REMOVED: dport->test_state = BFA_DPORT_ST_NO_SFP; break; case BFI_DPORT_SCN_DDPORT_DISABLE: bfa_fcport_ddportdisable(dport->bfa); if (bfa_dport_send_req(dport, BFI_DPORT_DYN_DISABLE)) bfa_sm_set_state(dport, bfa_dport_sm_dynamic_disabling); else bfa_sm_set_state(dport, bfa_dport_sm_dynamic_disabling_qwait); break; case BFI_DPORT_SCN_FCPORT_DISABLE: bfa_fcport_ddportdisable(dport->bfa); bfa_sm_set_state(dport, bfa_dport_sm_disabled); dport->dynamic = BFA_FALSE; break; default: bfa_trc(dport->bfa, dport->i2hmsg.scn.state); bfa_sm_fault(dport->bfa, event); } break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_disabling_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_QRESUME: bfa_sm_set_state(dport, bfa_dport_sm_disabling); bfa_dport_send_req(dport, BFI_DPORT_DISABLE); break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_reqq_wcancel(&dport->reqq_wait); bfa_cb_fcdiag_dport(dport, BFA_STATUS_OK); break; case BFA_DPORT_SM_SCN: /* ignore */ break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_disabling(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_FWRSP: dport->test_state = BFA_DPORT_ST_DISABLED; bfa_sm_set_state(dport, bfa_dport_sm_disabled); break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_cb_fcdiag_dport(dport, BFA_STATUS_OK); break; case BFA_DPORT_SM_SCN: /* no state change */ break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_starting_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_QRESUME: bfa_sm_set_state(dport, bfa_dport_sm_starting); bfa_dport_send_req(dport, BFI_DPORT_START); break; case BFA_DPORT_SM_HWFAIL: bfa_reqq_wcancel(&dport->reqq_wait); bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_cb_fcdiag_dport(dport, BFA_STATUS_FAILED); break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_starting(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_FWRSP: memset(&dport->result, 0, sizeof(struct bfa_diag_dport_result_s)); if (dport->i2hmsg.rsp.status == BFA_STATUS_DPORT_INV_SFP) { dport->test_state = BFA_DPORT_ST_NO_SFP; } else { dport->test_state = BFA_DPORT_ST_INP; bfa_dport_result_start(dport, BFA_DPORT_OPMODE_MANU); } /* fall thru */ case BFA_DPORT_SM_REQFAIL: bfa_sm_set_state(dport, bfa_dport_sm_enabled); break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_cb_fcdiag_dport(dport, BFA_STATUS_FAILED); break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_dynamic_disabling(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_SCN: switch (dport->i2hmsg.scn.state) { case BFI_DPORT_SCN_DDPORT_DISABLED: bfa_sm_set_state(dport, bfa_dport_sm_disabled); dport->dynamic = BFA_FALSE; bfa_fcport_enable(dport->bfa); break; default: bfa_trc(dport->bfa, dport->i2hmsg.scn.state); bfa_sm_fault(dport->bfa, event); } break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_cb_fcdiag_dport(dport, BFA_STATUS_OK); break; default: bfa_sm_fault(dport->bfa, event); } } static void bfa_dport_sm_dynamic_disabling_qwait(struct bfa_dport_s *dport, enum bfa_dport_sm_event event) { bfa_trc(dport->bfa, event); switch (event) { case BFA_DPORT_SM_QRESUME: bfa_sm_set_state(dport, bfa_dport_sm_dynamic_disabling); bfa_dport_send_req(dport, BFI_DPORT_DYN_DISABLE); break; case BFA_DPORT_SM_HWFAIL: bfa_sm_set_state(dport, bfa_dport_sm_disabled); bfa_reqq_wcancel(&dport->reqq_wait); bfa_cb_fcdiag_dport(dport, BFA_STATUS_OK); break; case BFA_DPORT_SM_SCN: /* ignore */ break; default: bfa_sm_fault(dport->bfa, event); } } static bfa_boolean_t bfa_dport_send_req(struct bfa_dport_s *dport, enum bfi_dport_req req) { struct bfi_diag_dport_req_s *m; /* * check for room in queue to send request now */ m = bfa_reqq_next(dport->bfa, BFA_REQQ_DIAG); if (!m) { bfa_reqq_wait(dport->bfa, BFA_REQQ_PORT, &dport->reqq_wait); return BFA_FALSE; } bfi_h2i_set(m->mh, BFI_MC_DIAG, BFI_DIAG_H2I_DPORT, bfa_fn_lpu(dport->bfa)); m->req = req; if ((req == BFI_DPORT_ENABLE) || (req == BFI_DPORT_START)) { m->lpcnt = cpu_to_be32(dport->lpcnt); m->payload = cpu_to_be32(dport->payload); } /* * queue I/O message to firmware */ bfa_reqq_produce(dport->bfa, BFA_REQQ_DIAG, m->mh); return BFA_TRUE; } static void bfa_dport_qresume(void *cbarg) { struct bfa_dport_s *dport = cbarg; bfa_sm_send_event(dport, BFA_DPORT_SM_QRESUME); } static void bfa_dport_req_comp(struct bfa_dport_s *dport, struct bfi_diag_dport_rsp_s *msg) { msg->status = cpu_to_be32(msg->status); dport->i2hmsg.rsp.status = msg->status; dport->rp_pwwn = msg->pwwn; dport->rp_nwwn = msg->nwwn; if ((msg->status == BFA_STATUS_OK) || (msg->status == BFA_STATUS_DPORT_NO_SFP)) { bfa_trc(dport->bfa, msg->status); bfa_trc(dport->bfa, dport->rp_pwwn); bfa_trc(dport->bfa, dport->rp_nwwn); bfa_sm_send_event(dport, BFA_DPORT_SM_FWRSP); } else { bfa_trc(dport->bfa, msg->status); bfa_sm_send_event(dport, BFA_DPORT_SM_REQFAIL); } bfa_cb_fcdiag_dport(dport, msg->status); } static bfa_boolean_t bfa_dport_is_sending_req(struct bfa_dport_s *dport) { if (bfa_sm_cmp_state(dport, bfa_dport_sm_enabling) || bfa_sm_cmp_state(dport, bfa_dport_sm_enabling_qwait) || bfa_sm_cmp_state(dport, bfa_dport_sm_disabling) || bfa_sm_cmp_state(dport, bfa_dport_sm_disabling_qwait) || bfa_sm_cmp_state(dport, bfa_dport_sm_starting) || bfa_sm_cmp_state(dport, bfa_dport_sm_starting_qwait)) { return BFA_TRUE; } else { return BFA_FALSE; } } static void bfa_dport_scn(struct bfa_dport_s *dport, struct bfi_diag_dport_scn_s *msg) { int i; uint8_t subtesttype; bfa_trc(dport->bfa, msg->state); dport->i2hmsg.scn.state = msg->state; switch (dport->i2hmsg.scn.state) { case BFI_DPORT_SCN_TESTCOMP: dport->result.end_time = bfa_get_log_time(); bfa_trc(dport->bfa, dport->result.end_time); dport->result.status = msg->info.testcomp.status; bfa_trc(dport->bfa, dport->result.status); dport->result.roundtrip_latency = cpu_to_be32(msg->info.testcomp.latency); dport->result.est_cable_distance = cpu_to_be32(msg->info.testcomp.distance); dport->result.buffer_required = be16_to_cpu(msg->info.testcomp.numbuffer); dport->result.frmsz = be16_to_cpu(msg->info.testcomp.frm_sz); dport->result.speed = msg->info.testcomp.speed; bfa_trc(dport->bfa, dport->result.roundtrip_latency); bfa_trc(dport->bfa, dport->result.est_cable_distance); bfa_trc(dport->bfa, dport->result.buffer_required); bfa_trc(dport->bfa, dport->result.frmsz); bfa_trc(dport->bfa, dport->result.speed); for (i = DPORT_TEST_ELOOP; i < DPORT_TEST_MAX; i++) { dport->result.subtest[i].status = msg->info.testcomp.subtest_status[i]; bfa_trc(dport->bfa, dport->result.subtest[i].status); } break; case BFI_DPORT_SCN_TESTSKIP: case BFI_DPORT_SCN_DDPORT_ENABLE: memset(&dport->result, 0, sizeof(struct bfa_diag_dport_result_s)); break; case BFI_DPORT_SCN_TESTSTART: memset(&dport->result, 0, sizeof(struct bfa_diag_dport_result_s)); dport->rp_pwwn = msg->info.teststart.pwwn; dport->rp_nwwn = msg->info.teststart.nwwn; dport->lpcnt = cpu_to_be32(msg->info.teststart.numfrm); bfa_dport_result_start(dport, msg->info.teststart.mode); break; case BFI_DPORT_SCN_SUBTESTSTART: subtesttype = msg->info.teststart.type; dport->result.subtest[subtesttype].start_time = bfa_get_log_time(); dport->result.subtest[subtesttype].status = DPORT_TEST_ST_INPRG; bfa_trc(dport->bfa, subtesttype); bfa_trc(dport->bfa, dport->result.subtest[subtesttype].start_time); break; case BFI_DPORT_SCN_SFP_REMOVED: case BFI_DPORT_SCN_DDPORT_DISABLED: case BFI_DPORT_SCN_DDPORT_DISABLE: case BFI_DPORT_SCN_FCPORT_DISABLE: dport->result.status = DPORT_TEST_ST_IDLE; break; default: bfa_sm_fault(dport->bfa, msg->state); } bfa_sm_send_event(dport, BFA_DPORT_SM_SCN); } /* * Dport enable * * @param[in] *bfa - bfa data struct */ bfa_status_t bfa_dport_enable(struct bfa_s *bfa, u32 lpcnt, u32 pat, bfa_cb_diag_t cbfn, void *cbarg) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); struct bfa_dport_s *dport = &fcdiag->dport; /* * Dport is not support in MEZZ card */ if (bfa_mfg_is_mezz(dport->bfa->ioc.attr->card_type)) { bfa_trc(dport->bfa, BFA_STATUS_PBC); return BFA_STATUS_CMD_NOTSUPP_MEZZ; } /* * Dport is supported in CT2 or above */ if (!(bfa_asic_id_ct2(dport->bfa->ioc.pcidev.device_id))) { bfa_trc(dport->bfa, dport->bfa->ioc.pcidev.device_id); return BFA_STATUS_FEATURE_NOT_SUPPORTED; } /* * Check to see if IOC is down */ if (!bfa_iocfc_is_operational(bfa)) return BFA_STATUS_IOC_NON_OP; /* if port is PBC disabled, return error */ if (bfa_fcport_is_pbcdisabled(bfa)) { bfa_trc(dport->bfa, BFA_STATUS_PBC); return BFA_STATUS_PBC; } /* * Check if port mode is FC port */ if (bfa_ioc_get_type(&bfa->ioc) != BFA_IOC_TYPE_FC) { bfa_trc(dport->bfa, bfa_ioc_get_type(&bfa->ioc)); return BFA_STATUS_CMD_NOTSUPP_CNA; } /* * Check if port is in LOOP mode */ if ((bfa_fcport_get_cfg_topology(bfa) == BFA_PORT_TOPOLOGY_LOOP) || (bfa_fcport_get_topology(bfa) == BFA_PORT_TOPOLOGY_LOOP)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_TOPOLOGY_LOOP; } /* * Check if port is TRUNK mode */ if (bfa_fcport_is_trunk_enabled(bfa)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_ERROR_TRUNK_ENABLED; } /* * Check if diag loopback is running */ if (bfa_fcdiag_lb_is_running(bfa)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_DIAG_BUSY; } /* * Check to see if port is disable or in dport state */ if ((bfa_fcport_is_disabled(bfa) == BFA_FALSE) && (bfa_fcport_is_dport(bfa) == BFA_FALSE)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_PORT_NOT_DISABLED; } /* * Check if dport is in dynamic mode */ if (dport->dynamic) return BFA_STATUS_DDPORT_ERR; /* * Check if dport is busy */ if (bfa_dport_is_sending_req(dport)) return BFA_STATUS_DEVBUSY; /* * Check if dport is already enabled */ if (bfa_sm_cmp_state(dport, bfa_dport_sm_enabled)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_DPORT_ENABLED; } bfa_trc(dport->bfa, lpcnt); bfa_trc(dport->bfa, pat); dport->lpcnt = (lpcnt) ? lpcnt : DPORT_ENABLE_LOOPCNT_DEFAULT; dport->payload = (pat) ? pat : LB_PATTERN_DEFAULT; dport->cbfn = cbfn; dport->cbarg = cbarg; bfa_sm_send_event(dport, BFA_DPORT_SM_ENABLE); return BFA_STATUS_OK; } /* * Dport disable * * @param[in] *bfa - bfa data struct */ bfa_status_t bfa_dport_disable(struct bfa_s *bfa, bfa_cb_diag_t cbfn, void *cbarg) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); struct bfa_dport_s *dport = &fcdiag->dport; if (bfa_ioc_is_disabled(&bfa->ioc)) return BFA_STATUS_IOC_DISABLED; /* if port is PBC disabled, return error */ if (bfa_fcport_is_pbcdisabled(bfa)) { bfa_trc(dport->bfa, BFA_STATUS_PBC); return BFA_STATUS_PBC; } /* * Check if dport is in dynamic mode */ if (dport->dynamic) { return BFA_STATUS_DDPORT_ERR; } /* * Check to see if port is disable or in dport state */ if ((bfa_fcport_is_disabled(bfa) == BFA_FALSE) && (bfa_fcport_is_dport(bfa) == BFA_FALSE)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_PORT_NOT_DISABLED; } /* * Check if dport is busy */ if (bfa_dport_is_sending_req(dport)) return BFA_STATUS_DEVBUSY; /* * Check if dport is already disabled */ if (bfa_sm_cmp_state(dport, bfa_dport_sm_disabled)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_DPORT_DISABLED; } dport->cbfn = cbfn; dport->cbarg = cbarg; bfa_sm_send_event(dport, BFA_DPORT_SM_DISABLE); return BFA_STATUS_OK; } /* * Dport start -- restart dport test * * @param[in] *bfa - bfa data struct */ bfa_status_t bfa_dport_start(struct bfa_s *bfa, u32 lpcnt, u32 pat, bfa_cb_diag_t cbfn, void *cbarg) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); struct bfa_dport_s *dport = &fcdiag->dport; /* * Check to see if IOC is down */ if (!bfa_iocfc_is_operational(bfa)) return BFA_STATUS_IOC_NON_OP; /* * Check if dport is in dynamic mode */ if (dport->dynamic) return BFA_STATUS_DDPORT_ERR; /* * Check if dport is busy */ if (bfa_dport_is_sending_req(dport)) return BFA_STATUS_DEVBUSY; /* * Check if dport is in enabled state. * Test can only be restart when previous test has completed */ if (!bfa_sm_cmp_state(dport, bfa_dport_sm_enabled)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_DPORT_DISABLED; } else { if (dport->test_state == BFA_DPORT_ST_NO_SFP) return BFA_STATUS_DPORT_INV_SFP; if (dport->test_state == BFA_DPORT_ST_INP) return BFA_STATUS_DEVBUSY; WARN_ON(dport->test_state != BFA_DPORT_ST_COMP); } bfa_trc(dport->bfa, lpcnt); bfa_trc(dport->bfa, pat); dport->lpcnt = (lpcnt) ? lpcnt : DPORT_ENABLE_LOOPCNT_DEFAULT; dport->payload = (pat) ? pat : LB_PATTERN_DEFAULT; dport->cbfn = cbfn; dport->cbarg = cbarg; bfa_sm_send_event(dport, BFA_DPORT_SM_START); return BFA_STATUS_OK; } /* * Dport show -- return dport test result * * @param[in] *bfa - bfa data struct */ bfa_status_t bfa_dport_show(struct bfa_s *bfa, struct bfa_diag_dport_result_s *result) { struct bfa_fcdiag_s *fcdiag = BFA_FCDIAG_MOD(bfa); struct bfa_dport_s *dport = &fcdiag->dport; /* * Check to see if IOC is down */ if (!bfa_iocfc_is_operational(bfa)) return BFA_STATUS_IOC_NON_OP; /* * Check if dport is busy */ if (bfa_dport_is_sending_req(dport)) return BFA_STATUS_DEVBUSY; /* * Check if dport is in enabled state. */ if (!bfa_sm_cmp_state(dport, bfa_dport_sm_enabled)) { bfa_trc(dport->bfa, 0); return BFA_STATUS_DPORT_DISABLED; } /* * Check if there is SFP */ if (dport->test_state == BFA_DPORT_ST_NO_SFP) return BFA_STATUS_DPORT_INV_SFP; memcpy(result, &dport->result, sizeof(struct bfa_diag_dport_result_s)); return BFA_STATUS_OK; }