PRD: created MemEcc::handleMemCe()

This is needed for the runtime command complete ECC analysis. Most
of the function already existed in a local function. So make the
function externally available.

Change-Id: Icd8b9a187f45a4b112a58a29c633b939d7a44cc0
RTC: 171915
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/39679
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Brian J. Stegmiller <bjs@us.ibm.com>
Reviewed-by: Caleb N. Palmer <cnpalmer@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/39900
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>

2 files changed, 86 insertions, 60 deletions

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
index 491b1f63b..b5e19080b 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
@@ -323,14 +323,16 @@ uint32_t analyzeFetchMpe<TYPE_MCA, McaDataBundle *>( ExtensibleChip * i_chip,
 //------------------------------------------------------------------------------
 
 template<TARGETING::TYPE T, typename D>
-uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
-                               const MemSymbol & i_symbol,
-                               STEP_CODE_DATA_STRUCT & io_sc )
+uint32_t handleMemCe( ExtensibleChip * i_chip, const MemAddr & i_addr,
+                      const MemSymbol & i_symbol, bool & o_doTps,
+                      STEP_CODE_DATA_STRUCT & io_sc, bool i_isHard )
 {
-    #define PRDF_FUNC "[MemEcc::__analyzeFetchNceTce] "
+    #define PRDF_FUNC "[MemEcc::handleMemCe] "
 
     uint32_t o_rc = SUCCESS;
 
+    o_doTps = i_isHard; // Do TPS on every hard CE.
+
     TargetHandle_t trgt = i_chip->getTrgt();
     MemRank        rank = i_addr.getRank();
 
@@ -340,8 +342,7 @@ uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
 
     // Add data to the CE table.
     D db = static_cast<D>(i_chip->getDataBundle());
-    uint32_t ceTableRc = db->iv_ceTable.addEntry( i_addr, i_symbol );
-    bool doTps = false;
+    uint32_t ceTableRc = db->iv_ceTable.addEntry( i_addr, i_symbol, i_isHard );
 
     // Check MNFG thresholds, if needed.
     if ( mfgMode() )
@@ -350,19 +351,19 @@ uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
         {
             io_sc.service_data->AddSignatureList( trgt, PRDFSIG_MnfgDramCte );
             io_sc.service_data->setServiceCall();
-            doTps = true;
+            o_doTps = true;
         }
         else if ( 0 != (MemCeTable<T>::MNFG_TH_RANK & ceTableRc) )
         {
             io_sc.service_data->AddSignatureList( trgt, PRDFSIG_MnfgRankCte );
             io_sc.service_data->setServiceCall();
-            doTps = true;
+            o_doTps = true;
         }
         else if ( 0 != (MemCeTable<T>::MNFG_TH_DIMM & ceTableRc) )
         {
             io_sc.service_data->AddSignatureList( trgt, PRDFSIG_MnfgDimmCte );
             io_sc.service_data->setServiceCall();
-            doTps = true;
+            o_doTps = true;
         }
         else if ( 0 != (MemCeTable<T>::TABLE_FULL & ceTableRc) )
         {
@@ -377,7 +378,7 @@ uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
             io_sc.service_data->SetCallout( all_mm, MRU_MEDA );
             io_sc.service_data->SetCallout( trgt,   MRU_MEDA );
             io_sc.service_data->setServiceCall();
-            doTps = true;
+            o_doTps = true;
         }
         else if ( 0 != (MemCeTable<T>::ENTRY_TH_REACHED & ceTableRc) )
         {
@@ -387,7 +388,7 @@ uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
             // has been met. This is a potential flooding issue. So make
             // the DIMM callout predictive.
             io_sc.service_data->setServiceCall();
-            doTps = true;
+            o_doTps = true;
         }
     }
     else // field thresholds
@@ -396,27 +397,68 @@ uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
         // thresholds because of the scaling due to DRAM side. Therefore, we
         // cannot simply trigger TPS on any threshold. The field and MNFG
         // thresholds must be handled separately.
-        doTps = ( 0 != (MemCeTable<T>::FIELD_TH_ALL & ceTableRc) );
+        if ( !o_doTps )
+            o_doTps = ( 0 != (MemCeTable<T>::FIELD_TH_ALL & ceTableRc) );
     }
 
-    // Initiate a TPS procedure, if needed.
-    if ( doTps )
-    {
-        #ifdef __HOSTBOOT_RUNTIME
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
 
-        // If a MNFG threshold has been reached (predictive callout), we
-        // will still try to start TPS just in case MNFG disables the
-        // termination policy.
+//------------------------------------------------------------------------------
+
+template<TARGETING::TYPE T, typename D>
+uint32_t __analyzeFetchNceTce( ExtensibleChip * i_chip, const MemAddr & i_addr,
+                               STEP_CODE_DATA_STRUCT & io_sc,
+                               bool i_isTce = false )
+{
+    #define PRDF_FUNC "[MemEcc::__analyzeFetchNceTce] "
+
+    uint32_t o_rc = SUCCESS;
 
-        o_rc = addTpsEvent<T,D>( i_chip, rank, io_sc );
+    do
+    {
+        // Get the symbol of the failure.
+        MemSymbol symbol;
+        o_rc = getMemReadSymbol<T>( i_chip, i_addr.getRank(), symbol, i_isTce );
         if ( SUCCESS != o_rc )
         {
-            PRDF_ERR( PRDF_FUNC "addTpsEvent(0x%08x, m%ds%d) failed",
-                      i_chip->getHuid(), rank.getMaster(), rank.getSlave() );
+            PRDF_ERR( PRDF_FUNC "getMemReadSymbol(0x%08x) failed",
+                      i_chip->getHuid() );
+            break;
         }
 
-        #endif
-    }
+        // Add the symbol to the callout list and CE table.
+        bool doTps;
+        o_rc = handleMemCe<T,D>( i_chip, i_addr, symbol, doTps, io_sc );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "handleMemCe(0x%08x) failed",
+                      i_chip->getHuid() );
+            break;
+        }
+
+        // Initiate a TPS procedure, if needed.
+        if ( doTps )
+        {
+            #ifdef __HOSTBOOT_RUNTIME
+
+            // If a MNFG threshold has been reached (predictive callout), we
+            // will still try to start TPS just in case MNFG disables the
+            // termination policy.
+
+            o_rc = addTpsEvent<T,D>( i_chip, i_addr.getRank(), io_sc );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "addTpsEvent(0x%08x) failed",
+                          i_chip->getHuid() );
+            }
+
+            #endif
+        }
+
+    } while (0);
 
     return o_rc;
 
@@ -447,20 +489,9 @@ uint32_t analyzeFetchNce( ExtensibleChip * i_chip,
                       i_chip->getHuid() );
             break;
         }
-        MemRank rank = addr.getRank();
-
-        // Get the symbol of the failure.
-        MemSymbol symbol;
-        o_rc = getMemReadSymbol<T>( i_chip, rank, symbol );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "getMemReadSymbol(0x%08x) failed",
-                      i_chip->getHuid() );
-            break;
-        }
 
         // Complete analysis.
-        o_rc = __analyzeFetchNceTce<T,D>( i_chip, addr, symbol, io_sc );
+        o_rc = __analyzeFetchNceTce<T,D>( i_chip, addr, io_sc );
         if ( SUCCESS != o_rc )
         {
             PRDF_ERR( PRDF_FUNC "__analyzeFetchNceTce(0x%08x) failed",
@@ -503,24 +534,13 @@ uint32_t analyzeFetchTce( ExtensibleChip * i_chip,
         o_rc = getMemReadAddr<T>( i_chip, MemAddr::READ_NCE_ADDR, addr );
         if ( SUCCESS != o_rc )
         {
-            PRDF_ERR( PRDF_FUNC "getMemReadAddr(0x%08x, READ_TCE_ADDR) failed",
-                      i_chip->getHuid() );
-            break;
-        }
-        MemRank rank = addr.getRank();
-
-        // Get the first symbol of the failure.
-        MemSymbol firstSymbol;
-        o_rc = getMemReadSymbol<T>( i_chip, rank, firstSymbol );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "first getMemReadSymbol(0x%08x) failed",
+            PRDF_ERR( PRDF_FUNC "getMemReadAddr(0x%08x) failed",
                       i_chip->getHuid() );
             break;
         }
 
         // Complete analysis for first symbol.
-        o_rc = __analyzeFetchNceTce<T,D>( i_chip, addr, firstSymbol, io_sc );
+        o_rc = __analyzeFetchNceTce<T,D>( i_chip, addr, io_sc );
         if ( SUCCESS != o_rc )
         {
             PRDF_ERR( PRDF_FUNC "first __analyzeFetchNceTce(0x%08x) failed",
@@ -528,18 +548,8 @@ uint32_t analyzeFetchTce( ExtensibleChip * i_chip,
             break;
         }
 
-        // Get the second symbol of the failure.
-        MemSymbol secondSymbol;
-        o_rc = getMemReadSymbol<T>( i_chip, rank, secondSymbol, true );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "second getMemReadSymbol(0x%08x, true) failed",
-                      i_chip->getHuid() );
-            break;
-        }
-
         // Complete analysis for second symbol.
-        o_rc = __analyzeFetchNceTce<T,D>( i_chip, addr, secondSymbol, io_sc );
+        o_rc = __analyzeFetchNceTce<T,D>( i_chip, addr, io_sc, true );
         if ( SUCCESS != o_rc )
         {
             PRDF_ERR( PRDF_FUNC "second __analyzeFetchNceTce(0x%08x) failed",
diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H
index 1ddc31d55..f46705ea3 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H
@@ -48,6 +48,22 @@ namespace MemEcc
 {
 
 /**
+ * @brief  Adds the given symbol to the callout list and CE table. Returns true
+ *         if TPS is required.
+ * @param  i_chip   MCA or MBA.
+ * @param  i_addr   Failed address.
+ * @param  i_symbol Failed symbol.
+ * @param  o_doTps  True if TPS is required. False otherwise.
+ * @param  io_sc    The step code data struct.
+ * @param  i_isHard True if this is a hard CE. Default is false.
+ * @return Non-SUCCESS if an interal function fails, SUCCESS otherwise.
+ */
+template<TARGETING::TYPE T, typename D>
+uint32_t handleMemCe( ExtensibleChip * i_chip, const MemAddr & i_addr,
+                      const MemSymbol & i_symbol, bool & o_doTps,
+                      STEP_CODE_DATA_STRUCT & io_sc, bool i_isHard = false );
+
+/**
  * @brief  Will check if the UE is a side-effect attention and make a callout
  *         appropriately.
  * @param  i_chip   MCA or MBA.