PRD: Redesigned placing marks during TPS

Change-Id: I1d24edc6b956270cd7d633780740354d51be1695
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/9322
Tested-by: Jenkins Server
Reviewed-by: Bilicon Patil <bilpatil@in.ibm.com>
Reviewed-by: Prem Shanker Jha <premjha2@in.ibm.com>
Reviewed-by: Christopher T. Phan <cphan@us.ibm.com>
Reviewed-by: Sachin Gupta <sgupta2m@in.ibm.com>
Reviewed-by: BENJAMIN J. WEISENBECK <bweisenb@us.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/9474

4 files changed, 68 insertions, 32 deletions

diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaExtraSig.H b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaExtraSig.H
index ffd220e60..d90f01c37 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaExtraSig.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaExtraSig.H
@@ -48,9 +48,11 @@ PRDR_ERROR_SIGNATURE(MaintNCE_CTE,   0xffff0016, "", "Maint SOFT/INTER CTE");
 PRDR_ERROR_SIGNATURE(VcmVerified,     0xffff0020, "", "VCM: verified");
 PRDR_ERROR_SIGNATURE(VcmFalseAlarm,   0xffff0021, "", "VCM: false alarm");
 PRDR_ERROR_SIGNATURE(VcmBadSpare,     0xffff0022, "", "VCM: bad DRAM spare");
-PRDR_ERROR_SIGNATURE(VcmMarksUnavail, 0xffff0023, "",
-                     "VCM: No more marks available");
-PRDR_ERROR_SIGNATURE(VcmFalseAlarmExceeded, 0xffff0024, "",
+PRDR_ERROR_SIGNATURE(VcmCmAndSpare,   0xffff0023, "",
+                     "VCM: Both spare and chip mark used");
+PRDR_ERROR_SIGNATURE(VcmCmAndSm,      0xffff0024, "",
+                     "VCM: Both symbol and chip mark used");
+PRDR_ERROR_SIGNATURE(VcmFalseAlarmExceeded, 0xffff0025, "",
                      "VCM: false alarm threshold exceeded");
 
 PRDR_ERROR_SIGNATURE(DsdDramSpared,   0xffff0030, "", "DSD: DRAM spared");
@@ -79,5 +81,7 @@ PRDR_ERROR_SIGNATURE(TpsSymbolMark, 0xffff0063, "", "TPS: symbol mark placed");
 PRDR_ERROR_SIGNATURE(TpsChipMark,   0xffff0064, "", "TPS: chip mark placed");
 PRDR_ERROR_SIGNATURE(TpsMarksUnavail, 0xffff0065, "",
                      "TPS: No more marks available");
+PRDR_ERROR_SIGNATURE(TpsCmAndSpare, 0xffff0066, "",
+                     "TPS: Both spare and chip mark used");
 
 #endif // __prdfCenMbaExtraSig_H
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_common.C b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_common.C
index 3601a6a8b..1116eda1f 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_common.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_common.C
@@ -517,14 +517,14 @@ int32_t CenMbaTdCtlrCommon::handleMCE_VCM2( STEP_CODE_DATA_STRUCT & io_sc )
                 else
                 {
                     // Chip mark and DRAM spare are both used.
-                    io_sc.service_data->SetErrorSig( PRDFSIG_VcmMarksUnavail );
+                    io_sc.service_data->SetErrorSig( PRDFSIG_VcmCmAndSpare );
                     io_sc.service_data->SetServiceCall();
                 }
             }
             else
             {
                 // Chip mark is in place and sparing is not possible.
-                setTdSignature( io_sc, PRDFSIG_VcmMarksUnavail );
+                setTdSignature( io_sc, PRDFSIG_VcmCmAndSpare );
                 io_sc.service_data->SetServiceCall();
             }
         }
@@ -535,7 +535,7 @@ int32_t CenMbaTdCtlrCommon::handleMCE_VCM2( STEP_CODE_DATA_STRUCT & io_sc )
             // predictive.
             if ( iv_mark.getSM().isValid() )
             {
-                setTdSignature( io_sc, PRDFSIG_VcmMarksUnavail );
+                setTdSignature( io_sc, PRDFSIG_VcmCmAndSm );
                 io_sc.service_data->SetServiceCall();
             }
         }
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.C b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.C
index 3b09433e8..7d2f1a638 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.C
@@ -55,14 +55,26 @@ static const char *mbsCeStatReg[][ CE_REGS_PER_MBA ] = {
 
 //------------------------------------------------------------------------------
 
+// Helper structs for collectCeStats()
+struct DramCount_t
+{
+    uint8_t totalCount; uint8_t symbolCount;
+    DramCount_t() : totalCount(0), symbolCount(0) {}
+};
+typedef std::map<uint32_t, DramCount_t> DramCountMap;
+
+//------------------------------------------------------------------------------
+
 int32_t collectCeStats( ExtensibleChip * i_mbaChip, const CenRank & i_rank,
-                        MaintSymbols & o_maintStats, CenSymbol & o_highestDram,
+                        MaintSymbols & o_maintStats, CenSymbol & o_chipMark,
                         uint8_t i_thr )
 {
     #define PRDF_FUNC "[MemUtils::collectCeStats] "
 
     int32_t o_rc = SUCCESS;
 
+    o_chipMark = CenSymbol(); // Initially invalid.
+
     do
     {
         if ( 0 == i_thr ) // Must be non-zero
@@ -87,7 +99,7 @@ int32_t collectCeStats( ExtensibleChip * i_mbaChip, const CenRank & i_rank,
             o_rc = FAIL; break;
         }
 
-        bool isX4 = isDramWidthX4(mbaTrgt);
+        const bool isX4 = isDramWidthX4(mbaTrgt);
 
         // Get the current spares on this rank.
         CenSymbol sp0, sp1, ecc;
@@ -99,8 +111,7 @@ int32_t collectCeStats( ExtensibleChip * i_mbaChip, const CenRank & i_rank,
         }
 
         // Use this map to keep track of the total counts per DRAM.
-        typedef std::map<uint32_t, uint32_t> DramCount;
-        DramCount dramCounts;
+        DramCountMap dramCounts;
 
         const char * reg_str = NULL;
         SCAN_COMM_REGISTER_CLASS * reg = NULL;
@@ -129,11 +140,15 @@ int32_t collectCeStats( ExtensibleChip * i_mbaChip, const CenRank & i_rank,
                 uint8_t dram = CenSymbol::symbol2Dram( sym, isX4 );
 
                 // Keep track of the total DRAM counts.
-                dramCounts[dram] += count;
+                dramCounts[dram].totalCount += count;
 
                 // Add any symbols that have exceeded threshold to the list.
                 if ( i_thr <= count )
                 {
+                    // Keep track of the total number of symbols per DRAM that
+                    // have exceeded threshold.
+                    dramCounts[dram].symbolCount++;
+
                     SymbolData symData;
                     symData.symbol = CenSymbol::fromSymbol( mbaTrgt, i_rank,
                                             sym, CenSymbol::BOTH_SYMBOL_DQS );
@@ -176,26 +191,35 @@ int32_t collectCeStats( ExtensibleChip * i_mbaChip, const CenRank & i_rank,
         std::sort( o_maintStats.begin(), o_maintStats.end(), sortSymDataCount );
 
         // Get the DRAM with the highest count.
-        DramCount::iterator highestEntry = dramCounts.begin();
-        DramCount::iterator it = highestEntry; ++it; // sets it to next entry
-        for ( ; it != dramCounts.end(); ++it )
+        uint32_t highestDram  = 0;
+        uint32_t highestCount = 0;
+        const uint32_t symbolTH = isX4 ? 1 : 2;
+        for ( DramCountMap::iterator it = dramCounts.begin();
+              it != dramCounts.end(); ++it )
         {
-            if ( highestEntry->second < it->second )
-                highestEntry = it;
+            if ( (symbolTH     <= it->second.symbolCount) &&
+                 (highestCount <  it->second.totalCount ) )
+            {
+                highestDram  = it->first;
+                highestCount = it->second.totalCount;
+            }
         }
 
-        uint8_t sym = CenSymbol::dram2Symbol( highestEntry->first, isX4 );
-        o_highestDram = CenSymbol::fromSymbol( mbaTrgt, i_rank, sym );
-
-        // Check if this symbol is on any of the spares.
-        if ( ( sp0.isValid() && (sp0.getDram() == o_highestDram.getDram()) ) ||
-             ( sp1.isValid() && (sp1.getDram() == o_highestDram.getDram()) ) )
-        {
-            o_highestDram.setDramSpared();
-        }
-        if ( ecc.isValid() && (ecc.getDram() == o_highestDram.getDram()) )
+        if ( 0 != highestCount )
         {
-            o_highestDram.setEccSpared();
+            uint8_t sym = CenSymbol::dram2Symbol( highestDram, isX4 );
+            o_chipMark  = CenSymbol::fromSymbol( mbaTrgt, i_rank, sym );
+
+            // Check if this symbol is on any of the spares.
+            if ( ( sp0.isValid() && (sp0.getDram() == o_chipMark.getDram()) ) ||
+                 ( sp1.isValid() && (sp1.getDram() == o_chipMark.getDram()) ) )
+            {
+                o_chipMark.setDramSpared();
+            }
+            if ( ecc.isValid() && (ecc.getDram() == o_chipMark.getDram()) )
+            {
+                o_chipMark.setEccSpared();
+            }
         }
 
     } while(0);
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.H b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.H
index b2a8292f2..cb4f6792c 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMemUtils.H
@@ -116,10 +116,18 @@ class MatchSymDataSymbol
  * @param  i_mbaChip     Target MBA chip.
  * @param  i_rank        Target rank.
  * @param  o_maintStats  Returns the list of symbols and counts.
- * @param  o_highestDram Returns a symbol representing the DRAM with the highest
- *                       total count. This includes counts from DRAM that are
- *                       under the given threshold. This value is undefined if
- *                       o_maintStats is empty.
+ * @param  o_chipMark    This represents the DRAM in which a chip mark should be
+ *                       placed, if necessary, based on the following:
+ *                         - In x4 mode, this represents the DRAM with the
+ *                           highest total per symbol count with at least 1
+ *                           symbol that has exceeded threshold.
+ *                         - In x8 mode, this represents the DRAM with the
+ *                           highest total per symbol count with at least 2
+ *                           symbols that have exceeded threshold.
+ *                       It is possible that there will be symbols returned in
+ *                       o_maintStats that do not meet the criteria stated
+ *                       above. Therefore, the user must check if o_chipMark is
+ *                       valid before using it.
  * @param  i_thr         The count threshold. Each symbol count must be greater
  *                       than or equal to this value to be added to the list.
  *                       The default is 1, which means all non-zero counts will
@@ -128,7 +136,7 @@ class MatchSymDataSymbol
  * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
  */
 int32_t collectCeStats( ExtensibleChip * i_mbaChip, const CenRank & i_rank,
-                        MaintSymbols & o_maintStats, CenSymbol & o_highestDram,
+                        MaintSymbols & o_maintStats, CenSymbol & o_chipMark,
                         uint8_t i_thr = 1 );
 
 /**