scsi: lpfc: Implement GID_PT on Nameserver query to support faster failover
authorJames Smart <jsmart2021@gmail.com>
Tue, 23 Oct 2018 20:41:10 +0000 (13:41 -0700)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 7 Nov 2018 01:42:51 +0000 (20:42 -0500)
The switches seem to respond faster to GID_PT vs GID_FT NameServer
queries.  Add support for GID_PT to be used over GID_FT to enable
faster storage failover detection. Includes addition of new module
parameter to select between GID_PT and GID_FT (GID_FT is default).

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/lpfc/lpfc.h
drivers/scsi/lpfc/lpfc_attr.c
drivers/scsi/lpfc/lpfc_crtn.h
drivers/scsi/lpfc/lpfc_ct.c
drivers/scsi/lpfc/lpfc_els.c
drivers/scsi/lpfc/lpfc_hbadisc.c
drivers/scsi/lpfc/lpfc_hw.h
drivers/scsi/lpfc/lpfc_hw4.h
drivers/scsi/lpfc/lpfc_nportdisc.c

index 95f0cdb..579c80b 100644 (file)
@@ -784,6 +784,7 @@ struct lpfc_hba {
 #define LPFC_FCF_PRIORITY 2    /* Priority fcf failover */
        uint32_t cfg_fcf_failover_policy;
        uint32_t cfg_fcp_io_sched;
+       uint32_t cfg_ns_query;
        uint32_t cfg_fcp2_no_tgt_reset;
        uint32_t cfg_cr_delay;
        uint32_t cfg_cr_count;
index dda7f45..1b19e8c 100644 (file)
@@ -5070,6 +5070,18 @@ LPFC_ATTR_RW(fcp_io_sched, LPFC_FCP_SCHED_ROUND_ROBIN,
             "issuing commands [0] - Round Robin, [1] - Current CPU");
 
 /*
+ * lpfc_ns_query: Determine algrithmn for NameServer queries after RSCN
+ * range is [0,1]. Default value is 0.
+ * For [0], GID_FT is used for NameServer queries after RSCN (default)
+ * For [1], GID_PT is used for NameServer queries after RSCN
+ *
+ */
+LPFC_ATTR_RW(ns_query, LPFC_NS_QUERY_GID_FT,
+            LPFC_NS_QUERY_GID_FT, LPFC_NS_QUERY_GID_PT,
+            "Determine algorithm NameServer queries after RSCN "
+            "[0] - GID_FT, [1] - GID_PT");
+
+/*
 # lpfc_fcp2_no_tgt_reset: Determine bus reset behavior
 # range is [0,1]. Default value is 0.
 # For [0], bus reset issues target reset to ALL devices
@@ -5514,6 +5526,7 @@ struct device_attribute *lpfc_hba_attrs[] = {
        &dev_attr_lpfc_scan_down,
        &dev_attr_lpfc_link_speed,
        &dev_attr_lpfc_fcp_io_sched,
+       &dev_attr_lpfc_ns_query,
        &dev_attr_lpfc_fcp2_no_tgt_reset,
        &dev_attr_lpfc_cr_delay,
        &dev_attr_lpfc_cr_count,
@@ -6564,6 +6577,7 @@ void
 lpfc_get_cfgparam(struct lpfc_hba *phba)
 {
        lpfc_fcp_io_sched_init(phba, lpfc_fcp_io_sched);
+       lpfc_ns_query_init(phba, lpfc_ns_query);
        lpfc_fcp2_no_tgt_reset_init(phba, lpfc_fcp2_no_tgt_reset);
        lpfc_cr_delay_init(phba, lpfc_cr_delay);
        lpfc_cr_count_init(phba, lpfc_cr_count);
index e9b297a..a4b1bc2 100644 (file)
@@ -175,6 +175,7 @@ void lpfc_hb_timeout_handler(struct lpfc_hba *);
 void lpfc_ct_unsol_event(struct lpfc_hba *, struct lpfc_sli_ring *,
                         struct lpfc_iocbq *);
 int lpfc_ct_handle_unsol_abort(struct lpfc_hba *, struct hbq_dmabuf *);
+int lpfc_issue_gidpt(struct lpfc_vport *vport);
 int lpfc_issue_gidft(struct lpfc_vport *vport);
 int lpfc_get_gidft_type(struct lpfc_vport *vport, struct lpfc_iocbq *iocbq);
 int lpfc_ns_cmd(struct lpfc_vport *, int, uint8_t, uint32_t);
index 789ad15..62e8ae3 100644 (file)
@@ -832,6 +832,198 @@ out:
 }
 
 static void
+lpfc_cmpl_ct_cmd_gid_pt(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
+                       struct lpfc_iocbq *rspiocb)
+{
+       struct lpfc_vport *vport = cmdiocb->vport;
+       struct Scsi_Host *shost = lpfc_shost_from_vport(vport);
+       IOCB_t *irsp;
+       struct lpfc_dmabuf *outp;
+       struct lpfc_dmabuf *inp;
+       struct lpfc_sli_ct_request *CTrsp;
+       struct lpfc_sli_ct_request *CTreq;
+       struct lpfc_nodelist *ndlp;
+       int rc;
+
+       /* First save ndlp, before we overwrite it */
+       ndlp = cmdiocb->context_un.ndlp;
+
+       /* we pass cmdiocb to state machine which needs rspiocb as well */
+       cmdiocb->context_un.rsp_iocb = rspiocb;
+       inp = (struct lpfc_dmabuf *)cmdiocb->context1;
+       outp = (struct lpfc_dmabuf *)cmdiocb->context2;
+       irsp = &rspiocb->iocb;
+
+       lpfc_debugfs_disc_trc(vport, LPFC_DISC_TRC_CT,
+                             "GID_PT cmpl:     status:x%x/x%x rtry:%d",
+                             irsp->ulpStatus, irsp->un.ulpWord[4],
+                             vport->fc_ns_retry);
+
+       /* Don't bother processing response if vport is being torn down. */
+       if (vport->load_flag & FC_UNLOADING) {
+               if (vport->fc_flag & FC_RSCN_MODE)
+                       lpfc_els_flush_rscn(vport);
+               goto out;
+       }
+
+       if (lpfc_els_chk_latt(vport)) {
+               lpfc_printf_vlog(vport, KERN_INFO, LOG_DISCOVERY,
+                                "4108 Link event during NS query\n");
+               if (vport->fc_flag & FC_RSCN_MODE)
+                       lpfc_els_flush_rscn(vport);
+               lpfc_vport_set_state(vport, FC_VPORT_FAILED);
+               goto out;
+       }
+       if (lpfc_error_lost_link(irsp)) {
+               lpfc_printf_vlog(vport, KERN_INFO, LOG_DISCOVERY,
+                                "4101 NS query failed due to link event\n");
+               if (vport->fc_flag & FC_RSCN_MODE)
+                       lpfc_els_flush_rscn(vport);
+               goto out;
+       }
+
+       spin_lock_irq(shost->host_lock);
+       if (vport->fc_flag & FC_RSCN_DEFERRED) {
+               vport->fc_flag &= ~FC_RSCN_DEFERRED;
+               spin_unlock_irq(shost->host_lock);
+
+               /* This is a GID_PT completing so the gidft_inp counter was
+                * incremented before the GID_PT was issued to the wire.
+                */
+               vport->gidft_inp--;
+
+               /*
+                * Skip processing the NS response
+                * Re-issue the NS cmd
+                */
+               lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
+                                "4102 Process Deferred RSCN Data: x%x x%x\n",
+                                vport->fc_flag, vport->fc_rscn_id_cnt);
+               lpfc_els_handle_rscn(vport);
+
+               goto out;
+       }
+       spin_unlock_irq(shost->host_lock);
+
+       if (irsp->ulpStatus) {
+               /* Check for retry */
+               if (vport->fc_ns_retry < LPFC_MAX_NS_RETRY) {
+                       if (irsp->ulpStatus != IOSTAT_LOCAL_REJECT ||
+                           (irsp->un.ulpWord[4] & IOERR_PARAM_MASK) !=
+                           IOERR_NO_RESOURCES)
+                               vport->fc_ns_retry++;
+
+                       /* CT command is being retried */
+                       vport->gidft_inp--;
+                       rc = lpfc_ns_cmd(vport, SLI_CTNS_GID_PT,
+                                        vport->fc_ns_retry, GID_PT_N_PORT);
+                       if (rc == 0)
+                               goto out;
+               }
+               if (vport->fc_flag & FC_RSCN_MODE)
+                       lpfc_els_flush_rscn(vport);
+               lpfc_vport_set_state(vport, FC_VPORT_FAILED);
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
+                                "4103 GID_FT Query error: 0x%x 0x%x\n",
+                                irsp->ulpStatus, vport->fc_ns_retry);
+       } else {
+               /* Good status, continue checking */
+               CTreq = (struct lpfc_sli_ct_request *)inp->virt;
+               CTrsp = (struct lpfc_sli_ct_request *)outp->virt;
+               if (CTrsp->CommandResponse.bits.CmdRsp ==
+                   cpu_to_be16(SLI_CT_RESPONSE_FS_ACC)) {
+                       lpfc_printf_vlog(vport, KERN_INFO, LOG_DISCOVERY,
+                                        "4105 NameServer Rsp Data: x%x x%x\n",
+                                        vport->fc_flag,
+                                        CTreq->un.gid.Fc4Type);
+
+                       lpfc_ns_rsp(vport,
+                                   outp,
+                                   CTreq->un.gid.Fc4Type,
+                                   (uint32_t)(irsp->un.genreq64.bdl.bdeSize));
+               } else if (CTrsp->CommandResponse.bits.CmdRsp ==
+                          be16_to_cpu(SLI_CT_RESPONSE_FS_RJT)) {
+                       /* NameServer Rsp Error */
+                       if ((CTrsp->ReasonCode == SLI_CT_UNABLE_TO_PERFORM_REQ)
+                           && (CTrsp->Explanation == SLI_CT_NO_FC4_TYPES)) {
+                               lpfc_printf_vlog(
+                                       vport, KERN_INFO, LOG_DISCOVERY,
+                                       "4106 No NameServer Entries "
+                                       "Data: x%x x%x x%x x%x\n",
+                                       CTrsp->CommandResponse.bits.CmdRsp,
+                                       (uint32_t)CTrsp->ReasonCode,
+                                       (uint32_t)CTrsp->Explanation,
+                                       vport->fc_flag);
+
+                               lpfc_debugfs_disc_trc(
+                               vport, LPFC_DISC_TRC_CT,
+                               "GID_PT no entry  cmd:x%x rsn:x%x exp:x%x",
+                               (uint32_t)CTrsp->CommandResponse.bits.CmdRsp,
+                               (uint32_t)CTrsp->ReasonCode,
+                               (uint32_t)CTrsp->Explanation);
+                       } else {
+                               lpfc_printf_vlog(
+                                       vport, KERN_INFO, LOG_DISCOVERY,
+                                       "4107 NameServer Rsp Error "
+                                       "Data: x%x x%x x%x x%x\n",
+                                       CTrsp->CommandResponse.bits.CmdRsp,
+                                       (uint32_t)CTrsp->ReasonCode,
+                                       (uint32_t)CTrsp->Explanation,
+                                       vport->fc_flag);
+
+                               lpfc_debugfs_disc_trc(
+                               vport, LPFC_DISC_TRC_CT,
+                               "GID_PT rsp err1  cmd:x%x rsn:x%x exp:x%x",
+                               (uint32_t)CTrsp->CommandResponse.bits.CmdRsp,
+                               (uint32_t)CTrsp->ReasonCode,
+                               (uint32_t)CTrsp->Explanation);
+                       }
+               } else {
+                       /* NameServer Rsp Error */
+                       lpfc_printf_vlog(vport, KERN_ERR, LOG_DISCOVERY,
+                                        "4109 NameServer Rsp Error "
+                                        "Data: x%x x%x x%x x%x\n",
+                                        CTrsp->CommandResponse.bits.CmdRsp,
+                                        (uint32_t)CTrsp->ReasonCode,
+                                        (uint32_t)CTrsp->Explanation,
+                                        vport->fc_flag);
+
+                       lpfc_debugfs_disc_trc(
+                               vport, LPFC_DISC_TRC_CT,
+                               "GID_PT rsp err2  cmd:x%x rsn:x%x exp:x%x",
+                               (uint32_t)CTrsp->CommandResponse.bits.CmdRsp,
+                               (uint32_t)CTrsp->ReasonCode,
+                               (uint32_t)CTrsp->Explanation);
+               }
+               vport->gidft_inp--;
+       }
+       /* Link up / RSCN discovery */
+       if ((vport->num_disc_nodes == 0) &&
+           (vport->gidft_inp == 0)) {
+               /*
+                * The driver has cycled through all Nports in the RSCN payload.
+                * Complete the handling by cleaning up and marking the
+                * current driver state.
+                */
+               if (vport->port_state >= LPFC_DISC_AUTH) {
+                       if (vport->fc_flag & FC_RSCN_MODE) {
+                               lpfc_els_flush_rscn(vport);
+                               spin_lock_irq(shost->host_lock);
+                               vport->fc_flag |= FC_RSCN_MODE; /* RSCN still */
+                               spin_unlock_irq(shost->host_lock);
+                       } else {
+                               lpfc_els_flush_rscn(vport);
+                       }
+               }
+
+               lpfc_disc_start(vport);
+       }
+out:
+       cmdiocb->context_un.ndlp = ndlp; /* Now restore ndlp for free */
+       lpfc_ct_free_iocb(phba, cmdiocb);
+}
+
+static void
 lpfc_cmpl_ct_cmd_gff_id(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
                        struct lpfc_iocbq *rspiocb)
 {
@@ -1365,6 +1557,8 @@ lpfc_ns_cmd(struct lpfc_vport *vport, int cmdcode,
        bpl->tus.f.bdeFlags = 0;
        if (cmdcode == SLI_CTNS_GID_FT)
                bpl->tus.f.bdeSize = GID_REQUEST_SZ;
+       else if (cmdcode == SLI_CTNS_GID_PT)
+               bpl->tus.f.bdeSize = GID_REQUEST_SZ;
        else if (cmdcode == SLI_CTNS_GFF_ID)
                bpl->tus.f.bdeSize = GFF_REQUEST_SZ;
        else if (cmdcode == SLI_CTNS_GFT_ID)
@@ -1405,6 +1599,18 @@ lpfc_ns_cmd(struct lpfc_vport *vport, int cmdcode,
                rsp_size = FC_MAX_NS_RSP;
                break;
 
+       case SLI_CTNS_GID_PT:
+               CtReq->CommandResponse.bits.CmdRsp =
+                   cpu_to_be16(SLI_CTNS_GID_PT);
+               CtReq->un.gid.PortType = context;
+
+               if (vport->port_state < LPFC_NS_QRY)
+                       vport->port_state = LPFC_NS_QRY;
+               lpfc_set_disctmo(vport);
+               cmpl = lpfc_cmpl_ct_cmd_gid_pt;
+               rsp_size = FC_MAX_NS_RSP;
+               break;
+
        case SLI_CTNS_GFF_ID:
                CtReq->CommandResponse.bits.CmdRsp =
                        cpu_to_be16(SLI_CTNS_GFF_ID);
index ebd6c72..6db426f 100644 (file)
@@ -6371,6 +6371,7 @@ int
 lpfc_els_handle_rscn(struct lpfc_vport *vport)
 {
        struct lpfc_nodelist *ndlp;
+       struct lpfc_hba  *phba = vport->phba;
 
        /* Ignore RSCN if the port is being torn down. */
        if (vport->load_flag & FC_UNLOADING) {
@@ -6399,8 +6400,15 @@ lpfc_els_handle_rscn(struct lpfc_vport *vport)
                 * flush the RSCN.  Otherwise, the outstanding requests
                 * need to complete.
                 */
-               if (lpfc_issue_gidft(vport) > 0)
+               if (phba->cfg_ns_query == LPFC_NS_QUERY_GID_FT) {
+                       if (lpfc_issue_gidft(vport) > 0)
+                               return 1;
+               } else if (phba->cfg_ns_query == LPFC_NS_QUERY_GID_PT) {
+                       if (lpfc_issue_gidpt(vport) > 0)
+                               return 1;
+               } else {
                        return 1;
+               }
        } else {
                /* Nameserver login in question.  Revalidate. */
                if (ndlp) {
index bfc4ac8..1723382 100644 (file)
@@ -3942,6 +3942,35 @@ lpfc_issue_gidft(struct lpfc_vport *vport)
        return vport->gidft_inp;
 }
 
+/**
+ * lpfc_issue_gidpt - issue a GID_PT for all N_Ports
+ * @vport: The virtual port for which this call is being executed.
+ *
+ * This routine will issue a GID_PT to get a list of all N_Ports
+ *
+ * Return value :
+ *   0 - Failure to issue a GID_PT
+ *   1 - GID_PT issued
+ **/
+int
+lpfc_issue_gidpt(struct lpfc_vport *vport)
+{
+       /* Good status, issue CT Request to NameServer */
+       if (lpfc_ns_cmd(vport, SLI_CTNS_GID_PT, 0, GID_PT_N_PORT)) {
+               /* Cannot issue NameServer FCP Query, so finish up
+                * discovery
+                */
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_SLI,
+                                "0606 %s Port TYPE %x %s\n",
+                                "Failed to issue GID_PT to ",
+                                GID_PT_N_PORT,
+                                "Finishing discovery.");
+               return 0;
+       }
+       vport->gidft_inp++;
+       return 1;
+}
+
 /*
  * This routine handles processing a NameServer REG_LOGIN mailbox
  * command upon completion. It is setup in the LPFC_MBOXQ
index 009aa0e..ec12270 100644 (file)
@@ -115,6 +115,7 @@ struct lpfc_sli_ct_request {
                uint32_t PortID;
                struct gid {
                        uint8_t PortType;       /* for GID_PT requests */
+#define GID_PT_N_PORT  1
                        uint8_t DomainScope;
                        uint8_t AreaScope;
                        uint8_t Fc4Type;        /* for GID_FT requests */
index bbd0a57..d3fde54 100644 (file)
@@ -197,6 +197,10 @@ struct lpfc_sli_intf {
 #define        LPFC_FCP_SCHED_ROUND_ROBIN      0
 #define        LPFC_FCP_SCHED_BY_CPU           1
 
+/* Algrithmns for NameServer Query after RSCN */
+#define LPFC_NS_QUERY_GID_FT   0
+#define LPFC_NS_QUERY_GID_PT   1
+
 /* Delay Multiplier constant */
 #define LPFC_DMULT_CONST       651042
 #define LPFC_DMULT_MAX         1023
index 6827ffe..7d5693c 100644 (file)
@@ -1775,9 +1775,16 @@ lpfc_cmpl_reglogin_reglogin_issue(struct lpfc_vport *vport,
                        ndlp->nlp_fc4_type |= NLP_FC4_FCP;
 
                } else if (ndlp->nlp_fc4_type == 0) {
-                       rc = lpfc_ns_cmd(vport, SLI_CTNS_GFT_ID,
-                                        0, ndlp->nlp_DID);
-                       return ndlp->nlp_state;
+                       /* If we are only configured for FCP, the driver
+                        * should just issue PRLI for FCP. Otherwise issue
+                        * GFT_ID to determine if remote port supports NVME.
+                        */
+                       if (phba->cfg_enable_fc4_type != LPFC_ENABLE_FCP) {
+                               rc = lpfc_ns_cmd(vport, SLI_CTNS_GFT_ID,
+                                                0, ndlp->nlp_DID);
+                               return ndlp->nlp_state;
+                       }
+                       ndlp->nlp_fc4_type = NLP_FC4_FCP;
                }
 
                ndlp->nlp_prev_state = NLP_STE_REG_LOGIN_ISSUE;