PRD: Runtime memory CE thresholds in MNFG

Change-Id: I9df24fc5162866ef066245307b43f7ddb4161774
CQ: SW262935
Backport: release-fips811
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/12041
Tested-by: Jenkins Server
Reviewed-by: Bilicon Patil <bilpatil@in.ibm.com>
Reviewed-by: Christopher T. Phan <cphan@us.ibm.com>
Reviewed-by: Sachin Gupta <sgupta2m@in.ibm.com>
Reviewed-by: Prem Shanker Jha <premjha2@in.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>
Reviewed-by: Zane Shelley <zshelle@us.ibm.com>
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/12219

4 files changed, 87 insertions, 44 deletions

diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.C b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.C
index 52808dafb..93df95117 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.C
@@ -5,7 +5,9 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2013,2014              */
+/* Contributors Listed Below - COPYRIGHT 2013,2014                        */
+/* [+] International Business Machines Corp.                              */
+/*                                                                        */
 /*                                                                        */
 /* Licensed under the Apache License, Version 2.0 (the "License");        */
 /* you may not use this file except in compliance with the License.       */
@@ -43,10 +45,10 @@ using namespace CE_TABLE;
 
 //------------------------------------------------------------------------------
 
-bool CenMbaCeTable::addEntry( const CenAddr & i_addr,
-                              const CenSymbol & i_symbol, bool i_isHard )
+uint32_t CenMbaCeTable::addEntry( const CenAddr & i_addr,
+                                  const CenSymbol & i_symbol, bool i_isHard )
 {
-    bool o_doTps = false;
+    uint32_t o_rc = NO_TH_REACHED;
 
     TableData data ( i_addr, i_symbol.getDram(), i_symbol.getDramPins(),
                      i_symbol.getPortSlct(), i_symbol.getWiringType(),
@@ -79,26 +81,22 @@ bool CenMbaCeTable::addEntry( const CenAddr & i_addr,
     // Check the new entry's count for single entry threshold.
     if ( TPS_ENTRY_COUNT_TH <= data.count )
     {
-        o_doTps = true;
+        o_rc |= ENTRY_TH_REACHED;
     }
 
     // Get number of entries in this table on the same rank as the new entry and
     // threshold if needed.
-    if ( !o_doTps ) // no point iterating the table if o_doTps is already true
+    CenRank thisRank = data.addr.getRank();
+    uint32_t rankCount = 0;
+    for ( CeTable::iterator it = iv_table.begin(); it != iv_table.end(); it++ )
     {
-        CenRank thisRank = data.addr.getRank();
-        uint32_t rankCount = 0;
-        for ( CeTable::iterator it = iv_table.begin();
-              it != iv_table.end(); it++ )
-        {
-            if ( it->active && (it->addr.getRank() == thisRank) )
-                rankCount++;
-        }
-
-        if ( TPS_RANK_ENTRY_TH <= rankCount )
-            o_doTps = true;
+        if ( it->active && (it->addr.getRank() == thisRank) )
+            rankCount++;
     }
 
+    if ( TPS_RANK_ENTRY_TH <= rankCount )
+        o_rc |= RANK_TH_REACHED;
+
     // Note that we intentially checked the entries-per-rank threshold before
     // removing any entries, if the table is full. This is to catch the corner
     // case where the oldest entry is on the same rank as the new entry.
@@ -120,13 +118,14 @@ bool CenMbaCeTable::addEntry( const CenAddr & i_addr,
     // If the table is still full, all entries are active. Pop off the oldest
     // active entry.
     if ( MAX_ENTRIES < iv_table.size() )
+    {
         iv_table.pop_front();
 
-    // Do TPS if the table is full.
-    if ( MAX_ENTRIES == iv_table.size() )
-        o_doTps = true;
+        // The table is full of active entries so do TPS.
+        o_rc |= TABLE_FULL;
+    }
 
-    return o_doTps;
+    return o_rc;
 }
 
 //------------------------------------------------------------------------------
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.H b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.H
index 8b0d7d29d..3842819ba 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaCeTable.H
@@ -5,7 +5,9 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2013,2014              */
+/* Contributors Listed Below - COPYRIGHT 2013,2014                        */
+/* [+] International Business Machines Corp.                              */
+/*                                                                        */
 /*                                                                        */
 /* Licensed under the Apache License, Version 2.0 (the "License");        */
 /* you may not use this file except in compliance with the License.       */
@@ -52,6 +54,17 @@ class CenMark;
 class CenMbaCeTable
 {
 
+  public: // constants, enums
+
+    /** @brief Return values from addEntry(). */
+    enum AddEntryRc
+    {
+        NO_TH_REACHED    = 0x00, ///< No thresholds reached.
+        ENTRY_TH_REACHED = 0x01, ///< The entry threshold has been reached.
+        RANK_TH_REACHED  = 0x02, ///< The rank threshold has been reached.
+        TABLE_FULL       = 0x04, ///< The table is full of active entries.
+    };
+
   private: // constants, enums
 
     /** @brief Table size limits. */
@@ -86,10 +99,10 @@ class CenMbaCeTable
      * @param  i_addr   The address reporting the CE.
      * @param  i_symbol The symbol reporting the CE.
      * @param  i_isHard TRUE if a hard CE was detected on this address/symbol.
-     * @return TRUE if TPS is required, FALSE otherwise.
+     * @return Mask of possible return codes (see enum AddEntryRc).
      */
-    bool addEntry( const CenAddr & i_addr, const CenSymbol & i_symbol,
-                   bool i_isHard = false );
+    uint32_t addEntry( const CenAddr & i_addr, const CenSymbol & i_symbol,
+                       bool i_isHard = false );
 
     /**
      * @brief Deactivates all entries in the table.
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembuf.C b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembuf.C
index ab75bad01..03e2aeb1c 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembuf.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembuf.C
@@ -560,15 +560,15 @@ int32_t AnalyzeFetchNce( ExtensibleChip * i_membChip,
 
             // Add the DIMM to the callout list
             MemoryMru memmru ( mbaTrgt, rank, symbol );
-            i_sc.service_data->SetCallout( memmru );
+            i_sc.service_data->SetCallout( memmru, MRU_MEDA );
 
             // Add to CE table
             CenMbaDataBundle * mbadb = getMbaDataBundle( mbaChip );
-            bool doTps = mbadb->iv_ceTable.addEntry( addr, symbol );
+            uint32_t ceTableRc = mbadb->iv_ceTable.addEntry( addr, symbol );
+            bool doTps = ( CenMbaCeTable::NO_TH_REACHED != ceTableRc );
 
-            // Check MNFG thresholds, if needed. No need to check if a TPS
-            // request is already needed.
-            if ( !doTps && mfgMode() )
+            // Check MNFG thresholds, if needed.
+            if ( mfgMode() )
             {
                 // Get the MNFG CE thresholds.
                 uint16_t dramTh, hrTh, dimmTh;
@@ -590,35 +590,61 @@ int32_t AnalyzeFetchNce( ExtensibleChip * i_membChip,
                 {
                     i_sc.service_data->AddSignatureList( mbaTrgt,
                                                          PRDFSIG_MnfgDramCte );
+                    i_sc.service_data->SetServiceCall();
                     doTps = true;
                 }
-
-                if ( hrTh < hrCount )
+                else if ( hrTh < hrCount )
                 {
                     i_sc.service_data->AddSignatureList( mbaTrgt,
                                                          PRDFSIG_MnfgHrCte );
+                    i_sc.service_data->SetServiceCall();
                     doTps = true;
                 }
-
-                if ( dimmTh < dimmCount )
+                else if ( dimmTh < dimmCount )
                 {
                     i_sc.service_data->AddSignatureList( mbaTrgt,
                                                          PRDFSIG_MnfgDimmCte );
+                    i_sc.service_data->SetServiceCall();
                     doTps = true;
                 }
+                else if ( 0 != (CenMbaCeTable::TABLE_FULL & ceTableRc) )
+                {
+                    i_sc.service_data->AddSignatureList( mbaTrgt,
+                                                         PRDFSIG_MnfgTableFull);
+
+                    // The table is full and no other threshold has been met.
+                    // We are in a state where we may never hit a MNFG
+                    // threshold. Callout all memory behind the MBA. Also, since
+                    // the counts are all over the place, there may be a problem
+                    // with the MBA. So call it out as well.
+                    MemoryMru all_mm ( mbaTrgt, rank,
+                                       MemoryMruData::CALLOUT_ALL_MEM );
+                    i_sc.service_data->SetCallout( all_mm,  MRU_MEDA );
+                    i_sc.service_data->SetCallout( mbaTrgt, MRU_MEDA );
+                    i_sc.service_data->SetServiceCall();
+                }
+                else if ( 0 != (CenMbaCeTable::ENTRY_TH_REACHED & ceTableRc) )
+                {
+                    i_sc.service_data->AddSignatureList( mbaTrgt,
+                                                         PRDFSIG_MnfgEntryCte );
+
+                    // There is a single entry threshold and no other threshold
+                    // has been met. This is a potential flooding issue, so make
+                    // the DIMM callout predictive.
+                    i_sc.service_data->SetServiceCall();
+                }
             }
 
             // Initiate a TPS procedure, if needed.
             if ( doTps )
             {
-                #ifdef __HOSTBOOT_MODULE
-                // Will not be able to do TPS during hostboot so make the error
-                // log predictive in MNFG mode. Note that we will still call
-                // handleTdEvent() so we can get the trace statement indicating
-                // TPS was requested during Hostboot.
-                if ( mfgMode() )
-                    i_sc.service_data->SetServiceCall();
-                #endif
+                // If a MNFG threshold has been reached (predictive callout), we
+                // will still try to start TPS just in case MNFG disables the
+                // termination policy.
+
+                // Will not be able to do TPS during hostboot. Note that we will
+                // still call handleTdEvent() so we can get the trace statement
+                // indicating TPS was requested during Hostboot.
 
                 l_rc = mbadb->iv_tdCtlr.handleTdEvent( i_sc, rank,
                                                 CenMbaTdCtlrCommon::TPS_EVENT );
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembufExtraSig.H b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembufExtraSig.H
index 862cc6c96..c409cbcdf 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembufExtraSig.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMembufExtraSig.H
@@ -5,7 +5,9 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2013,2014              */
+/* Contributors Listed Below - COPYRIGHT 2013,2014                        */
+/* [+] International Business Machines Corp.                              */
+/*                                                                        */
 /*                                                                        */
 /* Licensed under the Apache License, Version 2.0 (the "License");        */
 /* you may not use this file except in compliance with the License.       */
@@ -29,7 +31,10 @@
 PRDR_ERROR_SIGNATURE(MnfgDramCte,    0x44440000, "", "MNFG per DRAM CTE");
 PRDR_ERROR_SIGNATURE(MnfgHrCte,      0x44440001, "", "MNFG per half-rank CTE");
 PRDR_ERROR_SIGNATURE(MnfgDimmCte,    0x44440002, "", "MNFG per DIMM CTE");
-PRDR_ERROR_SIGNATURE(MbsFir_30_DD1Signature, 0x44440003, "MBSFIR[30]",
+PRDR_ERROR_SIGNATURE(MnfgTableFull,  0x44440003, "", "MNFG CE table full");
+PRDR_ERROR_SIGNATURE(MnfgEntryCte,   0x44440004, "", "MNFG CE table entry CTE");
+
+PRDR_ERROR_SIGNATURE(MbsFir_30_DD1Signature, 0x44440008, "MBSFIR[30]",
                                    "DD1: MBS_FIR_REG_INTERNAL_SCOM_ERROR_COPY");
 
 #endif // __prdfCenMembufExtraSig_H