PRD: cleanup after placing/verifying a chip mark on Centaur

Change-Id: I0688d00875ce97595a18b3338aff3f8f59e19ff3
RTC: 193261
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/59117
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Caleb N. Palmer <cnpalmer@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/gerrit1/59233
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>

9 files changed, 455 insertions, 336 deletions

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
index 467441c5a..bf0508d70 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
@@ -1111,25 +1111,14 @@ uint32_t analyzeImpe<TYPE_MCA>( ExtensibleChip * i_chip,
                     break;
                 }
 
-                o_rc = MarkStore::balance<TYPE_MCA>( i_chip, rank, io_sc );
+                o_rc = MarkStore::chipMarkCleanup<TYPE_MCA>( i_chip, rank,
+                                                             io_sc );
                 if ( SUCCESS != o_rc )
                 {
-                    PRDF_ERR( PRDF_FUNC "balance(0x%08x,0x%02x) failed",
+                    PRDF_ERR( PRDF_FUNC "chipMarkCleanup(0x%08x,0x%02x) failed",
                               i_chip->getHuid(), rank.getKey() );
                     break;
                 }
-
-                // Set the dram in DRAM Repairs VPD.
-                o_rc = setDramInVpd<TYPE_MCA>( i_chip, rank, symbol );
-                if ( SUCCESS != o_rc )
-                {
-                    PRDF_ERR( PRDF_FUNC "setDramInVpd(0x%08x,0x%02x) failed",
-                              i_chip->getHuid(), rank.getKey() );
-                    break;
-                }
-
-                // Add a DRAM sparing procedure to the queue, if supported.
-                // TODO: RTC 157888
             }
         }
 
diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemMark.C b/src/usr/diag/prdf/common/plat/mem/prdfMemMark.C
index 84e7c09b5..35f7803e7 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemMark.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemMark.C
@@ -30,6 +30,8 @@
 #include <prdfMemDbUtils.H>
 
 #ifdef __HOSTBOOT_MODULE
+#include <prdfCenMbaExtraSig.H>
+#include <prdfMemDsd.H>
 #include <prdfMemVcm.H>
 #endif
 
@@ -872,7 +874,290 @@ uint32_t writeSymbolMark<TYPE_MBA>( ExtensibleChip * i_chip,
     #undef PRDF_FUNC
 }
 
-//------------------------------------------------------------------------------
+//##############################################################################
+//          Utilities to cleanup markstore after a chip mark is verified
+//##############################################################################
+
+#ifdef __HOSTBOOT_MODULE // Not supported on FSP.
+
+template<TARGETING::TYPE T>
+uint32_t __applyRasPolicies( ExtensibleChip * i_chip, const MemRank & i_rank,
+                             STEP_CODE_DATA_STRUCT & io_sc,
+                             const MemMark & i_chipMark,
+                             const MemMark & i_symMark );
+
+template<>
+uint32_t __applyRasPolicies<TYPE_MCA>( ExtensibleChip * i_chip,
+                                       const MemRank & i_rank,
+                                       STEP_CODE_DATA_STRUCT & io_sc,
+                                       const MemMark & i_chipMark,
+                                       const MemMark & i_symMark )
+{
+    // There is no DRAM sparing on Nimbus so simply check if both the chip and
+    // symbol mark have been used.
+    if ( i_chipMark.isValid() && i_symMark.isValid() )
+    {
+        io_sc.service_data->setServiceCall();
+        io_sc.service_data->setSignature( i_chip->getHuid(),
+                                          PRDFSIG_AllDramRepairs );
+
+        #ifdef __HOSTBOOT_RUNTIME
+        // No more repairs left so no point doing any more TPS procedures.
+        MemDbUtils::banTps<TYPE_MCA>( i_chip, i_rank );
+        #endif
+    }
+
+    return SUCCESS;
+}
+
+template<>
+uint32_t __applyRasPolicies<TYPE_MBA>( ExtensibleChip * i_chip,
+                                       const MemRank & i_rank,
+                                       STEP_CODE_DATA_STRUCT & io_sc,
+                                       const MemMark & i_chipMark,
+                                       const MemMark & i_symMark )
+{
+    #define PRDF_FUNC "[__applyRasPolicies<TYPE_MBA>] "
+
+    uint32_t o_rc = SUCCESS;
+
+    bool allRepairsUsed = false;
+
+    do
+    {
+        const uint8_t ps   = i_chipMark.getSymbol().getPortSlct();
+        const uint8_t dram = i_chipMark.getSymbol().getDram();
+
+        const bool isX4 = isDramWidthX4( i_chip->getTrgt() );
+
+        // Determine if DRAM sparing is enabled.
+        bool isEnabled = isX4; // Always an ECC spare in x4 mode.
+
+        if ( !isEnabled )
+        {
+            /* TODO RTC 189221
+            // Check for any DRAM spares.
+            uint8_t cnfg = ENUM_ATTR_VPD_DIMM_SPARE_NO_SPARE;
+            o_rc = getDimmSpareConfig<TYPE_MBA>( i_chip, i_rank, ps, cnfg );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "getDimmSpareConfig(0x%08x,0x%02x,%d) "
+                          "failed", i_chip->getHuid(), i_rank.getKey(), ps );
+                break;
+            }
+            isEnabled = (ENUM_ATTR_VPD_DIMM_SPARE_NO_SPARE != cnfg);
+            */
+        }
+
+        if ( isEnabled )
+        {
+            // Sparing is enabled. Get the current spares in hardware.
+            MemSymbol sp0, sp1, ecc;
+            /* TODO RTC 189221
+            o_rc = mssGetSteerMux<TYPE_MBA>( i_chip, i_rank, sp0, sp1, ecc );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "mssGetSteerMux(0x%08x,0x%02x) failed",
+                          i_chip->getHuid(), i_rank.getKey() );
+                break;
+            }
+            */
+
+            // Add the spares to the callout list if they exist.
+            if ( sp0.isValid() )
+            {
+                MemoryMru mm { i_chip->getTrgt(), i_rank, sp0 };
+                io_sc.service_data->SetCallout( mm );
+            }
+            if ( sp1.isValid() )
+            {
+                MemoryMru mm { i_chip->getTrgt(), i_rank, sp1 };
+                io_sc.service_data->SetCallout( mm );
+            }
+            if ( ecc.isValid() )
+            {
+                MemoryMru mm { i_chip->getTrgt(), i_rank, ecc };
+                io_sc.service_data->SetCallout( mm );
+            }
+
+            // If the chip mark is on a spare then the spare is bad and hardware
+            // can not steer it to another DRAM even if one is available (e.g.
+            // the ECC spare). In this this case, make error log predictive.
+            if ( ( dram == (0 == ps ? sp0.getDram() : sp1.getDram()) ) ||
+                 ( dram == ecc.getDram() ) )
+            {
+                allRepairsUsed = true;
+                io_sc.service_data->setSignature( i_chip->getHuid(),
+                                                  PRDFSIG_VcmBadSpare );
+                break; // Nothing more to do.
+            }
+
+            // Certain DIMMs may have had spares intentially made unavailable by
+            // the manufacturer. Check the VPD for available spares.
+            bool dramSparePossible = false;
+            bool eccSparePossible  = false;
+            /* TODO RTC 189221
+            o_rc = bitmap.isSpareAvailable( ps, dramSparePossible,
+                                            eccSparePossible );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "isDramSpareAvailable() failed" );
+                break;
+            }
+            */
+
+            if ( dramSparePossible &&
+                 (0 == ps ? !sp0.isValid() : !sp1.isValid()) )
+            {
+                // A spare DRAM is available.
+                TdEntry * e = new DsdEvent<TYPE_MBA>{ i_chip, i_rank,
+                                                      i_chipMark };
+                MemDbUtils::pushToQueue<TYPE_MBA>( i_chip, e );
+            }
+            else if ( eccSparePossible && !ecc.isValid() )
+            {
+                // The ECC spare is available.
+                TdEntry * e = new DsdEvent<TYPE_MBA>{ i_chip, i_rank,
+                                                      i_chipMark, true };
+                MemDbUtils::pushToQueue<TYPE_MBA>( i_chip, e );
+            }
+            else
+            {
+                // Chip mark is in place and sparing is not possible.
+                allRepairsUsed = true;
+                io_sc.service_data->setSignature( i_chip->getHuid(),
+                                                  PRDFSIG_AllDramRepairs );
+            }
+        }
+        // There is no DRAM sparing so simply check if both the chip and symbol
+        // mark have been used.
+        else if ( i_chipMark.isValid() && i_symMark.isValid() )
+        {
+            allRepairsUsed = true;
+            io_sc.service_data->setSignature( i_chip->getHuid(),
+                                              PRDFSIG_AllDramRepairs );
+        }
+
+    } while (0);
+
+    if ( allRepairsUsed )
+    {
+        io_sc.service_data->setServiceCall();
+
+        #ifdef __HOSTBOOT_RUNTIME
+        // No more repairs left so no point doing any more TPS procedures.
+        MemDbUtils::banTps<TYPE_MCA>( i_chip, i_rank );
+        #endif
+    }
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+template<TARGETING::TYPE T>
+uint32_t chipMarkCleanup( ExtensibleChip * i_chip, const MemRank & i_rank,
+                          STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[chipMarkCleanup] "
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        // Get the chip mark.
+        MemMark chipMark;
+        o_rc = readChipMark<T>( i_chip, i_rank, chipMark );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "readChipMark(0x%08x,0x%02x) failed",
+                      i_chip->getHuid(), i_rank.getKey() );
+            break;
+        }
+
+        // There is nothing else to do if there is no chip mark.
+        if ( !chipMark.isValid() ) break;
+
+        // Add the chip mark to the callout list.
+        MemoryMru cm_mm { i_chip->getTrgt(), i_rank, chipMark.getSymbol() };
+        io_sc.service_data->SetCallout( cm_mm );
+
+        // Get the symbol mark.
+        MemMark symMark;
+        o_rc = readSymbolMark<T>( i_chip, i_rank, symMark );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "readSymbolMark(0x%08x,0x%02x) failed",
+                      i_chip->getHuid(), i_rank.getKey() );
+            break;
+        }
+
+        // If both the chip and symbol mark are on the same DRAM, clear the
+        // symbol mark.
+        if ( chipMark.getSymbol().getDram() == symMark.getSymbol().getDram() )
+        {
+            o_rc = clearSymbolMark<T>( i_chip, i_rank );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "clearSymbolMark(0x%08x,0x%02x) failed",
+                          i_chip->getHuid(), i_rank.getKey() );
+                break;
+            }
+
+            // Reset the symbol mark variable to invalid.
+            symMark = MemMark();
+        }
+
+        // Add the symbol mark to the callout list if it exists.
+        if ( symMark.isValid() )
+        {
+            MemoryMru sm_mm { i_chip->getTrgt(), i_rank, symMark.getSymbol() };
+            io_sc.service_data->SetCallout( sm_mm );
+        }
+
+        // Make the error log predictive and exit if DRAM repairs are disabled.
+        if ( areDramRepairsDisabled() )
+        {
+            io_sc.service_data->setServiceCall();
+            break; // nothing else to do
+        }
+
+        // Set the chip mark in the DRAM Repairs VPD.
+        o_rc = setDramInVpd<TYPE_MCA>( i_chip, i_rank, chipMark.getSymbol() );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "setDramInVpd(0x%08x,0x%02x) failed",
+                      i_chip->getHuid(), i_rank.getKey() );
+            break;
+        }
+
+        // Apply RAS policies.
+        o_rc = __applyRasPolicies<T>( i_chip, i_rank, io_sc, chipMark,
+                                      symMark );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "__applyRasPolicies(0x%08x,0x%02x) failed",
+                      i_chip->getHuid(), i_rank.getKey() );
+            break;
+        }
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+template
+uint32_t chipMarkCleanup<TYPE_MCA>( ExtensibleChip * i_chip,
+                                    const MemRank & i_rank,
+                                    STEP_CODE_DATA_STRUCT & io_sc );
+template
+uint32_t chipMarkCleanup<TYPE_MBA>( ExtensibleChip * i_chip,
+                                    const MemRank & i_rank,
+                                    STEP_CODE_DATA_STRUCT & io_sc );
+
+#endif // not supported on FSP
 
 } // end namespace MarkStore
 
diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemMark.H b/src/usr/diag/prdf/common/plat/mem/prdfMemMark.H
index 8ea692ea9..fca039258 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemMark.H
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemMark.H
@@ -174,62 +174,29 @@ template<TARGETING::TYPE T>
 uint32_t clearSymbolMark( ExtensibleChip * i_chip, const MemRank & i_rank );
 
 /**
- * @brief  If a rank contains a symbol mark that is on the same DRAM as the chip
- *         mark, the symbol mark is removed. This is done to free up available
- *         repairs. Will also apply RAS policies where necessary.
+ * @brief  If a chip mark has been verified or explicitly set due other RAS
+ *         policies, this function does all the necessary cleanup.
+ *
+ * Function details:
+ *  - Adds the chip mark to the callout list.
+ *  - Removes the symbol mark if it is on the same DRAM as the chip mark.
+ *  - Adds the symbol mark to the callout list if it exists on another DRAM.
+ *  - If DRAM repairs are disabled:
+ *      - Makes the error log predictive.
+ *  - Otherwise:
+ *      - Sets the DRAM in the DRAM Repair VPD.
+ *      - Makes the error log predictive if RAS policies apply.
+ *      - Adds a DSD procedure to the TD queue is a DRAM spare is available.
+ *      - Bans TPS on the rank if all repairs are used.
+ *
  * @param  i_chip MBA or MCA chip.
  * @param  i_rank Target rank.
  * @param  io_sc  The step code data struct.
  * @return Non-SUCCESS if an internal function fails. SUCCESS otherwise.
  */
 template<TARGETING::TYPE T>
-uint32_t balance( ExtensibleChip * i_chip, const MemRank & i_rank,
-                  STEP_CODE_DATA_STRUCT & io_sc )
-{
-    uint32_t o_rc = SUCCESS;
-
-    do
-    {
-        // Get the chip mark.
-        MemMark chipMark;
-        o_rc = readChipMark<T>( i_chip, i_rank, chipMark );
-        if ( SUCCESS != o_rc ) break;
-        if ( !chipMark.isValid() ) break; // nothing to do.
-
-        // Get the symbol mark.
-        MemMark symMark;
-        o_rc = readSymbolMark<T>( i_chip, i_rank, symMark );
-        if ( SUCCESS != o_rc ) break;
-        if ( !symMark.isValid() ) break; // nothing to do.
-
-        // If both the chip and symbol mark are on the same DRAM, clear the
-        // symbol mark.
-        if ( chipMark.getSymbol().getDram() == symMark.getSymbol().getDram() )
-        {
-            o_rc = clearSymbolMark<T>( i_chip, i_rank );
-            if ( SUCCESS != o_rc ) break;
-        }
-        else
-        {
-            // Both a chip and symbol mark exist, but they are on separate
-            // DRAMs. So, make the error log predictive.
-            io_sc.service_data->setServiceCall();
-            io_sc.service_data->setSignature( i_chip->getHuid(),
-                                              PRDFSIG_AllDramRepairs );
-
-            // The chip and symbol mark may be on different DIMMs (Centaur ranks
-            // span two DIMMs). Therefore, we must add both to the callout list
-            // to ensure all DIMMs are in the callout list.
-            MemoryMru cm_mm { i_chip->getTrgt(), i_rank, chipMark.getSymbol() };
-            MemoryMru sm_mm { i_chip->getTrgt(), i_rank, symMark.getSymbol()  };
-            io_sc.service_data->SetCallout( cm_mm );
-            io_sc.service_data->SetCallout( sm_mm );
-        }
-
-    } while (0);
-
-    return o_rc;
-}
+uint32_t chipMarkCleanup( ExtensibleChip * i_chip, const MemRank & i_rank,
+                          STEP_CODE_DATA_STRUCT & io_sc );
 
 } // end namespace MarkStore
 
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.C b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.C
index 88fd4dc00..f3ee6884f 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.C
@@ -596,87 +596,6 @@ int32_t CenMbaTdCtlr::startTpsPhase1( STEP_CODE_DATA_STRUCT & io_sc )
 
 //------------------------------------------------------------------------------
 
-int32_t CenMbaTdCtlr::addTdQueueEntryTPS( const CenRank & i_rank,
-                                          STEP_CODE_DATA_STRUCT & io_sc,
-                                          bool i_banTps )
-{
-    #define PRDF_FUNC "[CenMbaTdCtlr::addTdQueueEntryTPS] "
-
-    int32_t o_rc = SUCCESS;
-
-    do
-    {
-        if ( iv_tpsRankData.isBanned(i_rank, io_sc) )
-        {
-            // TPS is banned, do not add the request to the queue.
-            break;
-        }
-
-        // Check for any available repairs. There is no point doing TPS if we
-        // cannot apply a repair.
-        CenMark mark;
-        o_rc = mssGetMarkStore( iv_mbaTrgt, i_rank, mark );
-        if ( SUCCESS != o_rc )
-        {
-           PRDF_ERR( PRDF_FUNC "mssGetMarkStore() failed." );
-           break;
-        }
-        if ( mark.getCM().isValid() &&
-             (iv_x4Dimm || (!iv_x4Dimm && mark.getSM().isValid())) )
-        {
-            bool port0Available, port1Available;
-            o_rc  = checkForAvailableSpares( 0, port0Available );
-            o_rc |= checkForAvailableSpares( 1, port1Available );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "checkForAvailableSpares() failed." );
-                break;
-            }
-
-            if ( !port0Available && !port1Available )
-            {
-                // Ban TPS to avoid rechecking with subsequent TPS requests.
-                iv_tpsRankData.ban( iv_rank );
-
-                // TPS is banned, do not add the request to the queue.
-                break;
-            }
-        }
-
-        if ( i_banTps )
-        {
-            // Ban all future TPS requests for this rank (not including
-            // this one).
-            iv_tpsRankData.ban( i_rank );
-        }
-
-        // Push the TD request to the queue.
-        iv_queue.push( TdQueueEntry(TPS_EVENT, i_rank) );
-
-        // Mark this rank as bad.
-        o_rc = iv_masterRanks.setBad( i_rank );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "setBad() failed" );
-            break;
-        }
-
-    } while(0);
-
-    if ( SUCCESS != o_rc )
-    {
-        PRDF_ERR( PRDF_FUNC "Failed: i_rank=m%ds%d i_banTps=%c",
-                  i_rank.getMaster(), i_rank.getSlave(),
-                  i_banTps ? 'T' : 'F' );
-    }
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
 int32_t CenMbaTdCtlr::handleUe_Td( STEP_CODE_DATA_STRUCT & io_sc,
                                    const CenAddr & i_stopAddr,
                                    bool i_addTpsRequest )
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemDsd.H b/src/usr/diag/prdf/plat/mem/prdfMemDsd.H
index de1816927..73f133832 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemDsd.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemDsd.H
@@ -51,8 +51,9 @@ class DsdEvent : public TdEntry
      * @param i_rank Rank reporting chip mark.
      */
     DsdEvent<T>( ExtensibleChip * i_chip, const MemRank & i_rank,
-                 const MemMark & i_mark ) :
-        TdEntry(DSD_EVENT, i_chip, i_rank), iv_mark(i_mark)
+                 const MemMark & i_mark, bool i_eccSpare = false ) :
+        TdEntry(DSD_EVENT, i_chip, i_rank), iv_mark(i_mark),
+        iv_eccSpare(i_eccSpare)
     {
         PRDF_ASSERT( nullptr != i_chip );
         PRDF_ASSERT( T == i_chip->getType() );
@@ -166,7 +167,9 @@ class DsdEvent : public TdEntry
 
   private: // instance variables
 
-    const MemMark iv_mark; ///< The chip mark from hardware.
+    const MemMark iv_mark;  ///< The chip mark from hardware.
+    const bool iv_eccSpare; ///< True if the spare should be applied to the x4
+                            ///< DRAM ECC spare.
 };
 
 } // 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 57443c036..97cfeb557 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
@@ -947,16 +947,6 @@ uint32_t TpsEvent<TYPE_MCA>::analyzeCeSymbolCounts( CeCount i_badDqCount,
             }
         }
 
-        // We may have placed a chip mark, so if a symbol mark is being used on
-        // the same chip, undo the symbol mark after the chip mark is in place.
-        o_rc = MarkStore::balance<TYPE_MCA>( iv_chip, iv_rank, io_sc );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "MarkStore::balance(0x%08x,0x%02x) failed",
-                      iv_chip->getHuid(), getKey() );
-            break;
-        }
-
         // Write any updates to VPD.
         o_rc = setBadDqBitmap<DIMMS_PER_RANK::MCA>(mcaTrgt, iv_rank, dqBitmap);
         if ( SUCCESS != o_rc )
@@ -966,7 +956,19 @@ uint32_t TpsEvent<TYPE_MCA>::analyzeCeSymbolCounts( CeCount i_badDqCount,
                     iv_rank.getKey());
             break;
         }
-    }while(0);
+
+        // We may have placed a chip mark so do any necessary cleanup. This must
+        // be called after writing the bad DQ bitmap because the this function
+        // will also write it if necessary.
+        o_rc = MarkStore::chipMarkCleanup<TYPE_MCA>( iv_chip, iv_rank, io_sc );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "MarkStore::chipMarkCleanup(0x%08x,0x%02x) "
+                      "failed", iv_chip->getHuid(), getKey() );
+            break;
+        }
+
+    } while (0);
 
     return o_rc;
 
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm.H b/src/usr/diag/prdf/plat/mem/prdfMemVcm.H
index ccfa4475c..0de5f2fcf 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemVcm.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemVcm.H
@@ -277,19 +277,11 @@ class VcmEvent : public TdEntry
         io_sc.service_data->setSignature( iv_chip->getHuid(),
                                           PRDFSIG_VcmVerified );
 
-        if ( PlatServices::areDramRepairsDisabled() )
+        // Leave the chip mark in place and do any necessary cleanup.
+        o_rc = cleanup( io_sc );
+        if ( SUCCESS != o_rc )
         {
-            // Make the error log predictive, nothing else to do.
-            io_sc.service_data->setServiceCall();
-        }
-        else
-        {
-            // Leave the chip mark in place and do any necessary cleanup.
-            o_rc = cleanup( io_sc );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "cleanup() failed" );
-            }
+            PRDF_ERR( PRDF_FUNC "cleanup() failed" );
         }
 
         return o_rc;
@@ -305,7 +297,23 @@ class VcmEvent : public TdEntry
      * @param  io_sc  The step code data struct.
      * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
      */
-    uint32_t cleanup( STEP_CODE_DATA_STRUCT & io_sc );
+    uint32_t cleanup( STEP_CODE_DATA_STRUCT & io_sc )
+    {
+        #define PRDF_FUNC "[VcmEvent::cleanup] "
+
+        uint32_t o_rc = SUCCESS;
+
+        o_rc = MarkStore::chipMarkCleanup<T>( iv_chip, iv_rank, io_sc );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "chipMarkCleanup(0x%08x,0x%02x) failed",
+                    iv_chip->getHuid(), iv_rank.getKey() );
+        }
+
+        return o_rc;
+
+        #undef PRDF_FUNC
+    }
 
     /**
      * @brief  Verification failed. Do additional processing such as removing
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
index 69e8a26e6..d3de6d20d 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
@@ -83,46 +83,6 @@ uint32_t VcmEvent<T>::falseAlarm( STEP_CODE_DATA_STRUCT & io_sc )
 //------------------------------------------------------------------------------
 
 template<TARGETING::TYPE T>
-uint32_t VcmEvent<T>::cleanup( STEP_CODE_DATA_STRUCT & io_sc )
-{
-    #define PRDF_FUNC "[VcmEvent::cleanup] "
-
-    uint32_t o_rc = SUCCESS;
-
-    do
-    {
-        // If there is a symbol mark on the same DRAM as the newly verified chip
-        // mark, remove the symbol mark.
-        o_rc = MarkStore::balance<T>( iv_chip, iv_rank, io_sc );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "MarkStore::balance(0x%08x,0x%02x) failed",
-                      iv_chip->getHuid(), getKey() );
-            break;
-        }
-
-        // Set the dram in DRAM Repairs VPD.
-        o_rc = setDramInVpd<T>( iv_chip, iv_rank, iv_mark.getSymbol() );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "setDramInVpd(0x%08x,0x%02x) failed",
-                      iv_chip->getHuid(), iv_rank.getKey() );
-            break;
-        }
-
-        // Add a DRAM sparing procedure to the queue, if supported.
-        // TODO: RTC 157888
-
-    } while (0);
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
-template<TARGETING::TYPE T>
 bool __iueCheck( uint32_t i_eccAttns );
 
 template<> inline
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm_rt.C
index 64f760486..b7cb13653 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemVcm_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemVcm_rt.C
@@ -62,129 +62,6 @@ VcmFalseAlarm * __getFalseAlarmCounter<TYPE_MBA>( ExtensibleChip * i_chip )
 
 //##############################################################################
 //
-//                          Generic template functions
-//
-//##############################################################################
-
-template<TARGETING::TYPE T>
-uint32_t VcmEvent<T>::falseAlarm( STEP_CODE_DATA_STRUCT & io_sc )
-{
-    #define PRDF_FUNC "[VcmEvent::falseAlarm] "
-
-    uint32_t o_rc = SUCCESS;
-
-    PRDF_TRAC( PRDF_FUNC "Chip mark false alarm: 0x%08x,0x%02x",
-               iv_chip->getHuid(), getKey() );
-
-    io_sc.service_data->setSignature( iv_chip->getHuid(),
-                                      PRDFSIG_VcmFalseAlarm );
-
-    do
-    {
-        // If DRAM repairs are disabled, make the error log predictive.
-        if ( areDramRepairsDisabled() )
-        {
-            io_sc.service_data->setServiceCall();
-            break; // Nothing more to do.
-        }
-
-        // Increment the false alarm counter and check threshold.
-        uint8_t dram = iv_mark.getSymbol().getDram();
-        if ( __getFalseAlarmCounter<T>(iv_chip)->inc(iv_rank, dram, io_sc) )
-        {
-            // False alarm threshold has been reached.
-
-            io_sc.service_data->setSignature( iv_chip->getHuid(),
-                                              PRDFSIG_VcmFalseAlarmTH );
-
-            PRDF_TRAC( PRDF_FUNC "False alarm threshold: 0x%08x,0x%02x",
-                       iv_chip->getHuid(), getKey() );
-
-            // Leave the chip mark in place and do any necessary cleanup.
-            o_rc = cleanup( io_sc );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "cleanup() failed" );
-                break;
-            }
-        }
-        else
-        {
-            // Remove the chip mark.
-            o_rc = MarkStore::clearChipMark<T>( iv_chip, iv_rank );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "clearChipMark(0x%08x,0x%02x) failed",
-                          iv_chip->getHuid(), getKey() );
-                break;
-            }
-        }
-
-    } while (0);
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
-template<TARGETING::TYPE T>
-uint32_t VcmEvent<T>::cleanup( STEP_CODE_DATA_STRUCT & io_sc )
-{
-    #define PRDF_FUNC "[VcmEvent::cleanup] "
-
-    uint32_t o_rc = SUCCESS;
-
-    do
-    {
-        // If there is a symbol mark on the same DRAM as the newly verified chip
-        // mark, remove the symbol mark.
-        o_rc = MarkStore::balance<T>( iv_chip, iv_rank, io_sc );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "MarkStore::balance(0x%08x,0x%02x) failed",
-                      iv_chip->getHuid(), getKey() );
-            break;
-        }
-
-        // Set the dram in DRAM Repairs VPD.
-        o_rc = setDramInVpd<T>( iv_chip, iv_rank, iv_mark.getSymbol() );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "setDramInVpd(0x%08x,0x%02x) failed",
-                      iv_chip->getHuid(), iv_rank.getKey() );
-            break;
-        }
-
-        // Add a DRAM sparing procedure to the queue, if supported.
-        // TODO: RTC 157888
-
-        // The cleanup() function is called by both verified() and falseAlarm().
-        // In either case, we can pass in the DRAM characterized by iv_mark to
-        // determine if there has been a least one false alarm on any DRAM on
-        // this rank other than this DRAM. If so, the error log should be
-        // predictive.
-        VcmFalseAlarm * faCntr = __getFalseAlarmCounter<T>(iv_chip);
-        uint8_t dram = iv_mark.getSymbol().getDram();
-        if ( faCntr->queryDrams(iv_rank, dram, io_sc) )
-            io_sc.service_data->setServiceCall();
-
-    } while (0);
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
-// Avoid linker errors with the template.
-template class VcmEvent<TYPE_MCA>;
-template class VcmEvent<TYPE_MBA>;
-
-//##############################################################################
-//
 //                          Specializations for MCA
 //
 //##############################################################################
@@ -288,6 +165,42 @@ uint32_t VcmEvent<TYPE_MCA>::checkEcc( const uint32_t & i_eccAttns,
     #undef PRDF_FUNC
 }
 
+//------------------------------------------------------------------------------
+
+template<>
+uint32_t VcmEvent<TYPE_MCA>::cleanup( STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[VcmEvent::cleanup] "
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        o_rc = MarkStore::chipMarkCleanup<TYPE_MCA>( iv_chip, iv_rank, io_sc );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "chipMarkCleanup(0x%08x,0x%02x) failed",
+                      iv_chip->getHuid(), iv_rank.getKey() );
+            break;
+        }
+
+        // The cleanup() function is called by both verified() and falseAlarm().
+        // In either case, the error log should be predictive if there has been
+        // a least one false alarm on any DRAM on this rank other than this
+        // DRAM. This is required on Nimbus because of two symbol correction,
+        // which does not exist on Centaur.
+        VcmFalseAlarm * faCntr = __getFalseAlarmCounter<TYPE_MCA>(iv_chip);
+        uint8_t dram = iv_mark.getSymbol().getDram();
+        if ( faCntr->queryDrams(iv_rank, dram, io_sc) )
+            io_sc.service_data->setServiceCall();
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
 //##############################################################################
 //
 //                          Specializations for MBA
@@ -400,6 +313,79 @@ uint32_t VcmEvent<TYPE_MBA>::checkEcc( const uint32_t & i_eccAttns,
     #undef PRDF_FUNC
 }
 
+//##############################################################################
+//
+//                          Generic template functions
+//
+//##############################################################################
+
+template<TARGETING::TYPE T>
+uint32_t VcmEvent<T>::falseAlarm( STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[VcmEvent::falseAlarm] "
+
+    uint32_t o_rc = SUCCESS;
+
+    PRDF_TRAC( PRDF_FUNC "Chip mark false alarm: 0x%08x,0x%02x",
+               iv_chip->getHuid(), getKey() );
+
+    io_sc.service_data->setSignature( iv_chip->getHuid(),
+                                      PRDFSIG_VcmFalseAlarm );
+
+    do
+    {
+        // If DRAM repairs are disabled, make the error log predictive.
+        if ( areDramRepairsDisabled() )
+        {
+            io_sc.service_data->setServiceCall();
+            break; // Nothing more to do.
+        }
+
+        // Increment the false alarm counter and check threshold.
+        uint8_t dram = iv_mark.getSymbol().getDram();
+        if ( __getFalseAlarmCounter<T>(iv_chip)->inc(iv_rank, dram, io_sc) )
+        {
+            // False alarm threshold has been reached.
+
+            io_sc.service_data->setSignature( iv_chip->getHuid(),
+                                              PRDFSIG_VcmFalseAlarmTH );
+
+            PRDF_TRAC( PRDF_FUNC "False alarm threshold: 0x%08x,0x%02x",
+                       iv_chip->getHuid(), getKey() );
+
+            // Leave the chip mark in place and do any necessary cleanup.
+            o_rc = cleanup( io_sc );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "cleanup() failed" );
+                break;
+            }
+        }
+        else
+        {
+            // Remove the chip mark.
+            o_rc = MarkStore::clearChipMark<T>( iv_chip, iv_rank );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "clearChipMark(0x%08x,0x%02x) failed",
+                          iv_chip->getHuid(), getKey() );
+                break;
+            }
+        }
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
+// Avoid linker errors with the template.
+template class VcmEvent<TYPE_MCA>;
+template class VcmEvent<TYPE_MBA>;
+
 //------------------------------------------------------------------------------
 
 } // end namespace PRDF