Increase RTL to 500us

Stop waiting for APSS data when detected complete but failed

Add history counts for GPEs not idle and APSS collection errors

Change-Id: I2bfaf36ee0b736a958fd41c1c2145f537fc883c5
RTC: 172963
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/43052
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Reviewed-by: Andres A. Lugo-Reyes <aalugore@us.ibm.com>
Reviewed-by: William A. Bryan <wilbryan@us.ibm.com>
Reviewed-by: Christopher J. Cain <cjcain@us.ibm.com>

4 files changed, 44 insertions, 15 deletions

diff --git a/src/occ_405/dcom/dcomMasterRx.c b/src/occ_405/dcom/dcomMasterRx.c
index 2e26986..faa4dca 100644
--- a/src/occ_405/dcom/dcomMasterRx.c
+++ b/src/occ_405/dcom/dcomMasterRx.c
@@ -298,7 +298,7 @@ void task_dcom_rx_slv_outboxes( task_t *i_self)
                 else
                 {
                     L_bce_not_ready_count++;
-                    INCREMENT_ERR_HISTORY(ERR_DCOM_RX_SLV_OUTBOX);
+                    INCREMENT_ERR_HISTORY(ERRH_DCOM_RX_SLV_OUTBOX);
                 }
             }
             else
@@ -383,7 +383,7 @@ uint32_t dcom_rx_slv_outbox_doorbell( void )
 
     if (l_pbarc != 0)
     {
-        INCREMENT_ERR_HISTORY(ERR_DCOM_MASTER_PBAX_READ_FAIL);
+        INCREMENT_ERR_HISTORY(ERRH_DCOM_MASTER_PBAX_READ_FAIL);
 
         TRAC_ERR("Master PBAX Read Failure in receiving unicast slave doorbells - RC[%08X]", l_pbarc);
 
diff --git a/src/occ_405/dcom/dcomMasterTx.c b/src/occ_405/dcom/dcomMasterTx.c
index e1a46bc..a65dc31 100644
--- a/src/occ_405/dcom/dcomMasterTx.c
+++ b/src/occ_405/dcom/dcomMasterTx.c
@@ -89,7 +89,7 @@ uint32_t dcom_build_slv_inbox(void)
     // interrupt context.
     if(G_pbax_rc)
     {
-        INCREMENT_ERR_HISTORY(ERR_DCOM_MASTER_PBAX_SEND_FAIL);
+        INCREMENT_ERR_HISTORY(ERRH_DCOM_MASTER_PBAX_SEND_FAIL);
 
         if (!L_traced)
         {
@@ -232,6 +232,7 @@ void task_dcom_tx_slv_inbox( task_t *i_self)
     uint32_t    l_orc_ext = OCC_NO_EXTENDED_RC;
     uint64_t    l_start = ssx_timebase_get();
     bool        l_pwr_meas = FALSE;
+    bool        l_pwr_meas_complete_invalid = FALSE;
     bool        l_request_reset = FALSE;
     bool        l_ssx_failure = FALSE;
     // Use a static local bool to track whether the BCE request used
@@ -250,8 +251,9 @@ void task_dcom_tx_slv_inbox( task_t *i_self)
         }
 
         l_pwr_meas = G_ApssPwrMeasCompleted;
+        l_pwr_meas_complete_invalid = G_ApssPwrMeasDoneInvalid;
 
-        // Did APSS power complete?
+        // Did APSS power complete AND valid?
         if( l_pwr_meas == TRUE )
         {
 #if !defined(DEBUG_APSS_SEQ) && defined(DCOM_DEBUG)
@@ -405,13 +407,42 @@ void task_dcom_tx_slv_inbox( task_t *i_self)
             else
             {
                 L_bce_not_ready_count++;
-                INCREMENT_ERR_HISTORY(ERR_DCOM_TX_SLV_INBOX);
+                INCREMENT_ERR_HISTORY(ERRH_DCOM_TX_SLV_INBOX);
             }
 
             // Moved the break statement here in case we decide not to
             // schedule the BCE request.
             break;
         }
+        // Is APSS power collection done but failed?  Stop waiting since no valid data will be coming
+        else if(l_pwr_meas_complete_invalid == TRUE)
+        {
+            //Failure occurred, step up the FAIL_COUNT
+            APSS_FAIL();
+
+            if (G_apss_fail_updown_count >= APSS_DATA_FAIL_MAX)
+            {
+                if (FALSE == isSafeStateRequested())
+                {
+                        /* @
+                         * @errortype
+                         * @moduleid    DCOM_MID_TASK_TX_SLV_INBOX
+                         * @reasoncode  APSS_HARD_FAILURE
+                         * @userdata1   N/A
+                         * @userdata4   ERC_APSS_NO_VALID_DATA
+                         * @devdesc     No valid APSS data (hard time-out)
+                         */
+                        TRAC_ERR("No valid apss data: #complete errors = %i  #invalid = %i",
+                                 G_error_history[ERRH_APSS_COMPLETE_ERROR],
+                                 G_error_history[ERRH_INVALID_APSS_DATA]);
+                        l_orc = APSS_HARD_FAILURE;
+                        l_orc_ext = ERC_APSS_NO_VALID_DATA;
+                        l_request_reset = TRUE;
+                }
+            }
+
+            break;
+        }
         else
         {
             // check time and break out if we reached limit
@@ -421,7 +452,7 @@ void task_dcom_tx_slv_inbox( task_t *i_self)
             }
             else
             {
-                INCREMENT_ERR_HISTORY(ERR_DCOM_APSS_COMPLETE_TIMEOUT);
+                INCREMENT_ERR_HISTORY(ERRH_APSS_COMPLETE_TIMEOUT);
 
                 //Failure occurred, step up the FAIL_COUNT
                 APSS_FAIL();
@@ -430,9 +461,6 @@ void task_dcom_tx_slv_inbox( task_t *i_self)
                 {
                     if (FALSE == isSafeStateRequested())
                     {
-                        TRAC_ERR("task_dcom_tx_slv_inbox: APSS data collection failure exceeded threshold. fail_count=%i, threshold:%i",
-                                 G_apss_fail_updown_count, APSS_DATA_FAIL_MAX);
-
                         /* @
                          * @errortype
                          * @moduleid    DCOM_MID_TASK_TX_SLV_INBOX
@@ -450,6 +478,7 @@ void task_dcom_tx_slv_inbox( task_t *i_self)
                         l_request_reset = TRUE;
                     }
                 }
+
                 break;
             }
         }
diff --git a/src/occ_405/dcom/dcomSlaveRx.c b/src/occ_405/dcom/dcomSlaveRx.c
index a4b6e2d..7b103ab 100644
--- a/src/occ_405/dcom/dcomSlaveRx.c
+++ b/src/occ_405/dcom/dcomSlaveRx.c
@@ -278,7 +278,7 @@ void task_dcom_rx_slv_inbox( task_t *i_self)
                 else
                 {
                     L_bce_not_ready_count++;
-                    INCREMENT_ERR_HISTORY(ERR_DCOM_RX_SLV_INBOX);
+                    INCREMENT_ERR_HISTORY(ERRH_DCOM_RX_SLV_INBOX);
                 }
             }
             else
@@ -372,7 +372,7 @@ uint32_t dcom_rx_slv_inbox_doorbell( void )
         // We got an error reading from the PBAX, return to caller
         if ( l_pbarc != 0 )
         {
-            INCREMENT_ERR_HISTORY(ERR_DCOM_SLAVE_PBAX_READ_FAIL);
+            INCREMENT_ERR_HISTORY(ERRH_DCOM_SLAVE_PBAX_READ_FAIL);
             G_dcomTime.slave.doorbellErrorFlags |= DCOM_DOORBELL_HW_ERR;
             // Failure occurred
             TRAC_ERR("Slave PBAX Read Failure in receiving multicast doorbell from master - RC[%08X]", l_pbarc);
@@ -515,8 +515,8 @@ void task_dcom_wait_for_master( task_t *i_self)
 
                 if (L_no_master_doorbell_cnt == APSS_DATA_FAIL_PMAX_RAIL)
                 {
-                    // Now only trace every 50th occurrence
-                    L_trace_every_count = 50;
+                    // Now only trace every 1000th occurrence
+                    L_trace_every_count = 1000;
 
                     // Inform AMEC that Pmax_rail needs to change
                     G_apss_lower_pmax_rail = TRUE;
diff --git a/src/occ_405/dcom/dcomSlaveTx.c b/src/occ_405/dcom/dcomSlaveTx.c
index 33ef166..837608d 100755
--- a/src/occ_405/dcom/dcomSlaveTx.c
+++ b/src/occ_405/dcom/dcomSlaveTx.c
@@ -163,7 +163,7 @@ void task_dcom_tx_slv_outbox( task_t *i_self)
         if(G_slave_pbax_rc)
         {
             TRAC_ERR("task_dcom_tx_slv_outbox: PBAX Send Failure in transimitting doorbell - RC[%08X]", G_slave_pbax_rc);
-            INCREMENT_ERR_HISTORY(ERR_DCOM_SLAVE_PBAX_SEND_FAIL);
+            INCREMENT_ERR_HISTORY(ERRH_DCOM_SLAVE_PBAX_SEND_FAIL);
         }
 
         // Build/setup outbox
@@ -304,7 +304,7 @@ void task_dcom_tx_slv_outbox( task_t *i_self)
         else
         {
             L_bce_not_ready_count++;
-            INCREMENT_ERR_HISTORY(ERR_DCOM_TX_SLV_OUTBOX);
+            INCREMENT_ERR_HISTORY(ERRH_DCOM_TX_SLV_OUTBOX);
         }
 
     } while (0);