PRD: Runtime TPS Phase 0

Change-Id: I9aee499c606992705d3e0d3e8d660ba94efacd43
RTC: 171914
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/44217
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@us.ibm.com>
Reviewed-by: Brian J. Stegmiller <bjs@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/44764
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>

5 files changed, 167 insertions, 3 deletions

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfP9McaDataBundle.H b/src/usr/diag/prdf/common/plat/mem/prdfP9McaDataBundle.H
index 9ba08a2f1..11153b417 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfP9McaDataBundle.H
+++ b/src/usr/diag/prdf/common/plat/mem/prdfP9McaDataBundle.H
@@ -70,6 +70,7 @@ class McaDataBundle : public DataBundle
             delete iv_impeThresholdCounter;
             #ifdef __HOSTBOOT_RUNTIME
                 delete iv_vcmFalseAlarmCounter;
+                delete iv_tpsFalseAlarmCounter;
             #else
                 delete iv_iplCeStats;
             #endif
@@ -115,6 +116,18 @@ class McaDataBundle : public DataBundle
         return iv_vcmFalseAlarmCounter;
     }
 
+    /** @return The TPS false alarm counter. */
+    TpsFalseAlarm * getTpsFalseAlarmCounter()
+    {
+        if ( nullptr == iv_tpsFalseAlarmCounter )
+        {
+            iv_tpsFalseAlarmCounter = new TpsFalseAlarm(
+                TimeBasedThreshold{ 3, ThresholdResolution::ONE_DAY } );
+        }
+
+        return iv_tpsFalseAlarmCounter;
+    }
+
     #else // IPL only
 
     /** @return The IPL CE statistics object. */
@@ -152,6 +165,9 @@ class McaDataBundle : public DataBundle
     /** VCM false alarm counter. */
     VcmFalseAlarm * iv_vcmFalseAlarmCounter = nullptr;
 
+    /** TPS false alarm counter. */
+    TpsFalseAlarm * iv_tpsFalseAlarmCounter = nullptr;
+
     #else // IPL only
 
     /** MNFG IPL CE statistics. */
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTps.H b/src/usr/diag/prdf/plat/mem/prdfMemTps.H
index 6b05550cb..478cf9c47 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTps.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTps.H
@@ -59,6 +59,27 @@ class TpsEvent : public TdEntry
 
     uint32_t getKey() const { return iv_rank.getKey(); } // Master and slave
 
+  private: // functions
+
+    #ifdef __HOSTBOOT_RUNTIME
+
+    /**
+     * @brief  Starts phase 1 of the procedure.
+     * @param  io_sc  The step code data struct.
+     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+     */
+    uint32_t startTpsPhase1_rt( STEP_CODE_DATA_STRUCT & io_sc );
+
+    /**
+     * @brief  Analyzes phase 1 of the procedure.
+     * @param  io_sc  The step code data struct.
+     * @param  o_done True if the procedure is complete or has aborted.
+     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+     */
+    uint32_t analyzeTpsPhase1_rt(STEP_CODE_DATA_STRUCT & io_sc, bool & o_done);
+
+    #endif // __HOSTBOOT_RUNTIME
+
 };
 
 } // end namespace PRDF
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
index 13f054dca..d5af05261 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
@@ -26,7 +26,10 @@
 /** @file prdfMemTps_rt.C */
 
 // Platform includes
+#include <prdfMemTdFalseAlarm.H>
 #include <prdfMemTps.H>
+#include <prdfP9McaExtraSig.H>
+#include <prdfP9McaDataBundle.H>
 
 using namespace TARGETING;
 
@@ -35,6 +38,67 @@ namespace PRDF
 
 //------------------------------------------------------------------------------
 
+template <TARGETING::TYPE T>
+TpsFalseAlarm * __getTpsFalseAlarmCounter( ExtensibleChip * i_chip );
+
+template<>
+TpsFalseAlarm * __getTpsFalseAlarmCounter<TYPE_MCA>( ExtensibleChip * i_chip )
+{
+    return getMcaDataBundle(i_chip)->getTpsFalseAlarmCounter();
+}
+
+//------------------------------------------------------------------------------
+
+template<>
+TpsFalseAlarm * __getTpsFalseAlarmCounter<TYPE_MBA>( ExtensibleChip * i_chip )
+{
+    // TODO RTC 157888
+    //return getMbaDataBundle(i_chip)->getTpsFalseAlarmCounter();
+    return nullptr;
+}
+
+//------------------------------------------------------------------------------
+
+template<TARGETING::TYPE T>
+uint32_t TpsEvent<T>::startTpsPhase1_rt( STEP_CODE_DATA_STRUCT & io_sc )
+{
+    PRDF_TRAC( "[TpsEvent] Starting TPS Phase 1: 0x%08x,0x%02x",
+               iv_chip->getHuid(), getKey() );
+
+    iv_phase = TD_PHASE_1;
+    io_sc.service_data->AddSignatureList( iv_chip->getTrgt(),
+                                          PRDFSIG_StartTpsPhase1 );
+    bool countAllCes = false;
+    if ( __getTpsFalseAlarmCounter<T>(iv_chip)->count(iv_rank, io_sc) >= 1 )
+        countAllCes = true;
+
+    return PlatServices::startTpsRuntime<T>( iv_chip, iv_rank, countAllCes);
+}
+
+//------------------------------------------------------------------------------
+
+template<TARGETING::TYPE T>
+uint32_t TpsEvent<T>::analyzeTpsPhase1_rt( STEP_CODE_DATA_STRUCT & io_sc,
+                                           bool & o_done )
+{
+    #define PRDF_FUNC "[TpsEvent<T>::analyzeTpsPhase1_rt] "
+
+    uint32_t o_rc = SUCCESS;
+
+    // TODO RTC 171914
+    // Analyze Ecc Attentions
+    // Analyze CEs
+
+    o_done = true;
+    PRDF_ERR( PRDF_FUNC "function not implemented yet" );
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
 // TODO: RTC 171914 Actual implementation of this procedure will be done later.
 template<>
 uint32_t TpsEvent<TYPE_MCA>::nextStep( STEP_CODE_DATA_STRUCT & io_sc,
@@ -44,9 +108,27 @@ uint32_t TpsEvent<TYPE_MCA>::nextStep( STEP_CODE_DATA_STRUCT & io_sc,
 
     uint32_t o_rc = SUCCESS;
 
-    o_done = true;
-
-    PRDF_ERR( PRDF_FUNC "function not implemented yet" );
+    o_done = false;
+
+    switch ( iv_phase )
+    {
+        case TD_PHASE_0:
+            // Start TPS phase 1
+            o_rc = startTpsPhase1_rt( io_sc );
+            break;
+        case TD_PHASE_1:
+            // Analyze TPS phase 1
+            o_rc = analyzeTpsPhase1_rt( io_sc, o_done );
+            break;
+        default: PRDF_ASSERT( false ); // invalid phase
+
+    }
+
+    if ( SUCCESS != o_rc )
+    {
+        PRDF_ERR( PRDF_FUNC "TPS failed: 0x%08x,0x%02x", iv_chip->getHuid(),
+                  getKey() );
+    }
 
     return o_rc;
 
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
index 230bde80f..771ad402b 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -527,6 +527,27 @@ uint32_t startTpsPhase2<TYPE_MCA>( ExtensibleChip * i_mcaChip,
     return __startTdScrubSlave_mca( i_mcaChip, i_rank, stopCond );
 }
 
+//------------------------------------------------------------------------------
+
+template<>
+uint32_t startTpsRuntime<TYPE_MCA>( ExtensibleChip * i_mcaChip,
+                                    const MemRank & i_rank,
+                                    bool i_countAllCes )
+{
+    mss::mcbist::stop_conditions stopCond;
+    stopCond.set_nce_hard_symbol_count_enable(mss::ON);
+
+    // If the TPS false alarms count is one or more, enable per-symbol counters
+    // for soft and intermittent CEs.
+    if ( i_countAllCes )
+    {
+        stopCond.set_nce_soft_symbol_count_enable(mss::ON)
+                .set_nce_inter_symbol_count_enable(mss::ON);
+    }
+
+    return __startTdScrubSlave_mca( i_mcaChip, i_rank, stopCond );
+}
+
 //##############################################################################
 //##                   Centaur Maintenance Command wrappers
 //##############################################################################
@@ -583,6 +604,19 @@ uint32_t startTpsPhase2<TYPE_MBA>( ExtensibleChip * i_mbaChip,
     return SUCCESS;
 }
 
+//------------------------------------------------------------------------------
+
+template<>
+uint32_t startTpsRuntime<TYPE_MBA>( ExtensibleChip * i_mbaChip,
+                                    const MemRank & i_rank,
+                                    bool i_countAllCes )
+{
+    PRDF_ERR( "function not implemented yet" ); // TODO RTC 157888
+    return SUCCESS;
+}
+
+//------------------------------------------------------------------------------
+
 int32_t restartTraceArray(TargetHandle_t i_tgt)
 {
     int32_t o_rc = SUCCESS;
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.H b/src/usr/diag/prdf/plat/prdfPlatServices.H
index f4522c424..0c19c2e45 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.H
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.H
@@ -154,6 +154,17 @@ template<TARGETING::TYPE T>
 uint32_t startTpsPhase2( ExtensibleChip * i_chip, const MemRank & i_rank );
 
 /**
+ * @brief  Starts a targeted scrub command on the target rank for runtime TPS.
+ * @param  i_chip MCA or MBA chip.
+ * @param  i_rank The rank to target.
+ * @param  i_countAllCes True if we should count all CEs.
+ * @return Non-SUCCESS if an internal function fails, otherwise SUCCESS.
+ */
+template<TARGETING::TYPE T>
+uint32_t startTpsRuntime( ExtensibleChip * i_chip, const MemRank & i_rank,
+                          bool i_countAllCes );
+
+/**
  * @brief Restarts Trace arrays after having stopped on error
  * @param i_tgt EC/EQ target
  * @return non-SUCCESS for failure, SUCCESS otherwise