PRD: Maintenance RCE handling during background scrub for Centaur

Change-Id: Ib146e7bead1f3b4bae4e36fd582360bdbd22afce
RTC: 192638
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/58979
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Caleb N. Palmer <cnpalmer@us.ibm.com>
Reviewed-by: Matt Derksen <mderkse1@us.ibm.com>
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/59228
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>

3 files changed, 53 insertions, 210 deletions

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 4f8df05fa..24d7b6c9e 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.C
@@ -532,36 +532,6 @@ int32_t CenMbaTdCtlr::analyzeCmdComplete( STEP_CODE_DATA_STRUCT & io_sc,
             }
         }
 
-        if ( eccErrorMask & RETRY_CTE )
-        {
-            o_rc = handleRceEte_NonTd( io_sc );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "handleRceEte_NonTd() failed" );
-                break;
-            }
-        }
-
-        if ( eccErrorMask & MPE )
-        {
-            o_rc = handleMpe_NonTd( io_sc, i_stopAddr );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "handleMpe_NonTd() failed" );
-                break;
-            }
-        }
-
-        if ( eccErrorMask & UE )
-        {
-            o_rc = handleUe_NonTd( io_sc, i_stopAddr );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "handleUe_NonTd() failed" );
-                break;
-            }
-        }
-
         if ( iv_queue.empty() )
         {
             // No TD requests so resume background. If the scrub reached the end
@@ -1956,152 +1926,6 @@ int32_t CenMbaTdCtlr::handleCeEte_Tps( STEP_CODE_DATA_STRUCT & io_sc )
 
 //------------------------------------------------------------------------------
 
-int32_t CenMbaTdCtlr::handleUe_NonTd( STEP_CODE_DATA_STRUCT & io_sc,
-                                      const CenAddr & i_addr )
-{
-    #define PRDF_FUNC "[CenMbaTdCtlr::handleUe_NonTd] "
-
-    int32_t o_rc = SUCCESS;
-
-    setTdSignature( io_sc, PRDFSIG_MaintUE );
-
-    do
-    {
-        // Add entry to UE table.
-        CenMbaDataBundle * mbadb = getMbaDataBundle( iv_mbaChip );
-        mbadb->iv_ueTable.addEntry( UE_TABLE::SCRUB_UE, i_addr );
-
-        // Callout the rank.
-        MemoryMru memmru ( iv_mbaTrgt, iv_rank, MemoryMruData::CALLOUT_RANK );
-        io_sc.service_data->SetCallout( memmru );
-        io_sc.service_data->setServiceCall();
-
-        // Add a TPS request to the queue and ban any future TPS requests.
-        o_rc = addTdQueueEntryTPS( iv_rank, io_sc, true );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "addTdQueueEntryTPS() failed" );
-            break;
-        }
-
-        // Send lmb gard message to PHYP.
-        o_rc =  DEALLOC::lmbGard( iv_mbaChip, i_addr, false );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "lmbGard() failed" );
-            break;
-        }
-
-    } while(0);
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
-int32_t CenMbaTdCtlr::handleMpe_NonTd( STEP_CODE_DATA_STRUCT & io_sc,
-                                       const CenAddr & i_addr )
-{
-    #define PRDF_FUNC "[CenMbaTdCtlr::handleMpe_NonTd] "
-
-    int32_t o_rc = SUCCESS;
-
-    setTdSignature( io_sc, PRDFSIG_MaintMPE );
-
-    do
-    {
-        // Add entry to UE table.
-        CenMbaDataBundle * mbadb = getMbaDataBundle( iv_mbaChip );
-        mbadb->iv_ueTable.addEntry( UE_TABLE::SCRUB_MPE, i_addr );
-
-        // Add a VCM request to the queue.
-        o_rc = addTdQueueEntryVCM( iv_rank );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "addTdQueueEntryVCM() failed" );
-            break;
-        }
-
-        // Get the current mark in hardware.
-        CenMark mark;
-        o_rc = mssGetMarkStore( iv_mbaTrgt, iv_rank, mark );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "mssGetMarkStore() failed");
-            break;
-        }
-
-        // Callout the mark.
-        CalloutUtil::calloutMark( iv_mbaTrgt, iv_rank, mark, io_sc );
-
-    } while( 0 );
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
-int32_t CenMbaTdCtlr::handleRceEte_NonTd( STEP_CODE_DATA_STRUCT & io_sc )
-{
-    #define PRDF_FUNC "[CenMbaTdCtlr::handleRceEte_NonTd] "
-
-    int32_t o_rc = SUCCESS;
-
-    setTdSignature( io_sc, PRDFSIG_MaintRETRY_CTE );
-
-    do
-    {
-        MemoryMru memmru ( iv_mbaTrgt, iv_rank, MemoryMruData::CALLOUT_RANK );
-        io_sc.service_data->SetCallout( memmru );
-
-        bool doTps = true;
-
-        if ( mfgMode() )
-        {
-            // Get RCE count.
-            const char * reg_str = (0 == iv_mbaPos) ? "MBA0_MBSEC1"
-                                                    : "MBA1_MBSEC1";
-            SCAN_COMM_REGISTER_CLASS * mbsec1
-                                        = iv_membChip->getRegister( reg_str );
-            o_rc = mbsec1->Read();
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "Read() failed on %s", reg_str );
-                break;
-            }
-
-            uint16_t count = mbsec1->GetBitFieldJustified( 0, 12 );
-
-            // Add count to RCE table
-            CenMbaDataBundle * mbadb = getMbaDataBundle( iv_mbaChip );
-            doTps = mbadb->iv_rceTable.addEntry( iv_rank, io_sc, count );
-        }
-        else
-            io_sc.service_data->setServiceCall();
-
-        if ( doTps )
-        {
-            o_rc = addTdQueueEntryTPS( iv_rank, io_sc );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "addTdQueueEntryTPS() failed" );
-                break;
-            }
-        }
-
-    } while(0);
-
-    return o_rc;
-
-    #undef PRDF_FUNC
-}
-
-//------------------------------------------------------------------------------
-
 int32_t CenMbaTdCtlr::handleHardCeEte_NonTd( STEP_CODE_DATA_STRUCT & io_sc,
                                              const CenAddr & i_addr )
 {
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.H b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.H
index 40f67247d..2838d288a 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfCenMbaTdCtlr_rt.H
@@ -262,31 +262,6 @@ class CenMbaTdCtlr : public CenMbaTdCtlrCommon
     int32_t handleCeEte_Tps( STEP_CODE_DATA_STRUCT & io_sc );
 
     /**
-     * @brief  Handles UEs during background scrub.
-     * @param  io_sc  The step code data struct.
-     * @param  i_addr The address in which the maintenance command stopped.
-     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
-     */
-    int32_t handleUe_NonTd( STEP_CODE_DATA_STRUCT & io_sc,
-                            const CenAddr & i_addr );
-
-    /**
-     * @brief  Handles MPEs during background scrub.
-     * @param  io_sc  The step code data struct.
-     * @param  i_addr The address in which the maintenance command stopped.
-     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
-     */
-    int32_t handleMpe_NonTd( STEP_CODE_DATA_STRUCT & io_sc,
-                             const CenAddr & i_addr );
-
-    /**
-     * @brief  Handles RCE ETEs during background scrub.
-     * @param  io_sc The step code data struct.
-     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
-     */
-    int32_t handleRceEte_NonTd( STEP_CODE_DATA_STRUCT & io_sc );
-
-    /**
      * @brief  Handles hard CE ETEs during background scrub.
      * @param  io_sc  The step code data struct.
      * @param  i_addr The address in which the maintenance command stopped.
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
index e0835df7e..b42daa446 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
@@ -524,12 +524,13 @@ uint32_t __handleNceEte( ExtensibleChip * i_chip, TdQueue & io_queue,
 //------------------------------------------------------------------------------
 
 template<TARGETING::TYPE T>
-uint32_t __handleRceEte( ExtensibleChip * i_chip, bool & o_errorsFound,
+uint32_t __handleRceEte( ExtensibleChip * i_chip, TdQueue & io_queue,
+                         const MemRank & i_rank, bool & o_errorsFound,
                          STEP_CODE_DATA_STRUCT & io_sc );
 
 template<>
-uint32_t __handleRceEte<TYPE_MCA>( ExtensibleChip * i_chip,
-                                   bool & o_errorsFound,
+uint32_t __handleRceEte<TYPE_MCA>( ExtensibleChip * i_chip, TdQueue & io_queue,
+                                   const MemRank & i_rank, bool & o_errorsFound,
                                    STEP_CODE_DATA_STRUCT & io_sc )
 {
     #define PRDF_FUNC "[__handleRceEte] "
@@ -557,7 +558,7 @@ uint32_t __handleRceEte<TYPE_MCA>( ExtensibleChip * i_chip,
         o_errorsFound = true;
         io_sc.service_data->AddSignatureList( i_chip->getTrgt(),
                                               PRDFSIG_MaintIUE );
-        o_rc = MemEcc::analyzeMaintIue<TYPE_MCA>(i_chip, io_sc);
+        o_rc = MemEcc::handleMemIue<TYPE_MCA>( i_chip, i_rank, io_sc );
         if ( SUCCESS != o_rc )
         {
             PRDF_ERR( PRDF_FUNC "analyzeMaintIue(0x%08x) failed",
@@ -572,18 +573,61 @@ uint32_t __handleRceEte<TYPE_MCA>( ExtensibleChip * i_chip,
     #undef PRDF_FUNC
 }
 
-/* TODO RTC 157888
 template<>
-uint32_t __handleRceEte<TYPE_MBA>( ExtensibleChip * i_chip,
-                                   bool & o_errorsFound,
+uint32_t __handleRceEte<TYPE_MBA>( ExtensibleChip * i_chip, TdQueue & io_queue,
+                                   const MemRank & i_rank, bool & o_errorsFound,
                                    STEP_CODE_DATA_STRUCT & io_sc )
 {
     #define PRDF_FUNC "[__handleRceEte] "
 
     uint32_t o_rc = SUCCESS;
 
+    TargetHandle_t trgt = i_chip->getTrgt();
+
+    o_errorsFound = true;
+    io_sc.service_data->AddSignatureList( trgt, PRDFSIG_MaintRETRY_CTE );
+
+    // Add the rank to the callout list.
+    MemoryMru mm { trgt, i_rank, MemoryMruData::CALLOUT_RANK };
+    io_sc.service_data->SetCallout( mm );
+
     do
     {
+        bool doTps = true;
+
+        if ( mfgMode() )
+        {
+            ExtensibleChip * membChip = getConnectedParent(i_chip, TYPE_MEMBUF);
+
+            // Get the current RCE count from hardware.
+            const char * reg_str = (0 == i_chip->getPos()) ? "MBA0_MBSEC1"
+                                                           : "MBA1_MBSEC1";
+            SCAN_COMM_REGISTER_CLASS * reg = membChip->getRegister( reg_str );
+            o_rc = reg->Read();
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "Read() failed on %s", reg_str );
+                break;
+            }
+            uint16_t count = reg->GetBitFieldJustified( 0, 12 );
+
+            // Add the count to RCE table.
+            doTps = getMbaDataBundle(i_chip)->iv_rceTable.addEntry( i_rank,
+                                                                    io_sc,
+                                                                    count );
+        }
+        else
+        {
+            // The RCE threshold was set to the maximum. If we hit this then
+            // there is definitely a problem.
+            io_sc.service_data->setServiceCall();
+        }
+
+        // Add a TPS procedure to the queue, if needed.
+        if ( doTps )
+        {
+            io_queue.push( new TpsEvent<TYPE_MBA>(i_chip, i_rank) );
+        }
 
     } while (0);
 
@@ -591,7 +635,6 @@ uint32_t __handleRceEte<TYPE_MBA>( ExtensibleChip * i_chip,
 
     #undef PRDF_FUNC
 }
-*/
 
 //------------------------------------------------------------------------------
 
@@ -694,7 +737,8 @@ uint32_t __checkEcc( ExtensibleChip * i_chip, TdQueue & io_queue,
 
         if ( 0 != (eccAttns & MAINT_RCE_ETE) )
         {
-            o_rc = __handleRceEte<T>( i_chip, o_errorsFound, io_sc );
+            o_rc = __handleRceEte<T>( i_chip, io_queue, rank, o_errorsFound,
+                                      io_sc );
             if ( SUCCESS != o_rc )
             {
                 PRDF_ERR( PRDF_FUNC "__handleRceEte<T>(0x%08x) failed", huid );