PRD: TPS Analyze ECC attentions

Change-Id: I2326623087bf8d8b1c6d3a4d79472837e5c5857e
RTC: 171914
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/44368
Tested-by: Jenkins Server <pfd-jenkins+hostboot@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/gerrit1/44765
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>

4 files changed, 152 insertions, 40 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
index 40bb9df83..64719bf5f 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
@@ -181,28 +181,13 @@ uint32_t __checkEcc( ExtensibleChip * i_chip, TdQueue & io_queue,
         {
             io_sc.service_data->AddSignatureList( trgt, PRDFSIG_MaintMPE );
 
-            // Read the chip mark from markstore.
-            MemMark chipMark;
-            o_rc = MarkStore::readChipMark<T>( i_chip, rank, chipMark );
+            o_rc = MemEcc::handleMpe<T,D>( i_chip, rank, io_sc );
             if ( SUCCESS != o_rc )
             {
-                PRDF_ERR( PRDF_FUNC "readChipMark<T>(0x%08x,%d) failed",
-                          huid, rank.getMaster() );
+                PRDF_ERR( PRDF_FUNC "handleMpe<T>(0x%08x, 0x%02x) failed",
+                          i_chip->getHuid(), rank.getKey() );
                 break;
             }
-
-            // If the chip mark is not valid, then somehow the chip mark was
-            // placed on a rank other than the rank in which the command
-            // stopped. This would most likely be a code bug.
-            PRDF_ASSERT( chipMark.isValid() );
-
-            // Add the mark to the callout list.
-            MemoryMru mm { trgt, rank, chipMark.getSymbol() };
-            io_sc.service_data->SetCallout( mm );
-
-            // Add a new VCM procedure to the queue.
-            TdEntry * e = new VcmEvent<T>{ i_chip, rank, chipMark };
-            io_queue.push( e );
         }
         else if ( isMfgCeCheckingEnabled() &&
                   (0 != (eccAttns & MAINT_HARD_NCE_ETE)) )
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
index 069cd9ca6..8741c183d 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
@@ -664,28 +664,13 @@ uint32_t __checkEcc( ExtensibleChip * i_chip, TdQueue & io_queue,
             D db = static_cast<D>(i_chip->getDataBundle());
             db->iv_ueTable.addEntry( UE_TABLE::SCRUB_MPE, i_addr );
 
-            // Read the chip mark from markstore.
-            MemMark chipMark;
-            o_rc = MarkStore::readChipMark<T>( i_chip, rank, chipMark );
+            o_rc = MemEcc::handleMpe<T,D>( i_chip, rank, io_sc );
             if ( SUCCESS != o_rc )
             {
-                PRDF_ERR( PRDF_FUNC "readChipMark<T>(0x%08x,%d) failed",
-                          huid, rank.getMaster() );
+                PRDF_ERR( PRDF_FUNC "handleMpe<T>(0x%08x, 0x%02x) failed",
+                          i_chip->getHuid(), rank.getKey() );
                 break;
             }
-
-            // If the chip mark is not valid, then somehow the chip mark was
-            // placed on a rank other than the rank in which the command
-            // stopped. This would most likely be a code bug.
-            PRDF_ASSERT( chipMark.isValid() );
-
-            // Add the mark to the callout list.
-            MemoryMru mm { trgt, rank, chipMark.getSymbol() };
-            io_sc.service_data->SetCallout( mm );
-
-            // Add a VCM procedure to the queue.
-            TdEntry * e = new VcmEvent<T>{ i_chip, rank, chipMark };
-            io_queue.push( e );
         }
 
         if ( 0 != (eccAttns & MAINT_RCE_ETE) )
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTps.H b/src/usr/diag/prdf/plat/mem/prdfMemTps.H
index 478cf9c47..360cb55c5 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTps.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTps.H
@@ -78,6 +78,18 @@ class TpsEvent : public TdEntry
      */
     uint32_t analyzeTpsPhase1_rt(STEP_CODE_DATA_STRUCT & io_sc, bool & o_done);
 
+    /**
+     * @brief  Does isolation for ECC attentions.
+     * @param  i_eccAttns Mask of all currently active maintenance attentions.
+     *                    See enum MaintEccAttns for values.
+     * @param  io_sc      The step code data struct.
+     * @param  o_done     True if the procedure is complete or has aborted.
+     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+     */
+    uint32_t analyzeEcc( const uint32_t & i_eccAttns,
+                         STEP_CODE_DATA_STRUCT & io_sc,
+                         bool & o_done );
+
     #endif // __HOSTBOOT_RUNTIME
 
 };
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
index d5af05261..c0172a9cd 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTps_rt.C
@@ -26,16 +26,21 @@
 /** @file prdfMemTps_rt.C */
 
 // Platform includes
+#include <prdfMemEccAnalysis.H>
+#include <prdfMemScrubUtils.H>
 #include <prdfMemTdFalseAlarm.H>
 #include <prdfMemTps.H>
 #include <prdfP9McaExtraSig.H>
 #include <prdfP9McaDataBundle.H>
+#include <prdfTargetServices.H>
 
 using namespace TARGETING;
 
 namespace PRDF
 {
 
+using namespace PlatServices;
+
 //------------------------------------------------------------------------------
 
 template <TARGETING::TYPE T>
@@ -86,11 +91,32 @@ uint32_t TpsEvent<T>::analyzeTpsPhase1_rt( STEP_CODE_DATA_STRUCT & io_sc,
     uint32_t o_rc = SUCCESS;
 
     // TODO RTC 171914
-    // Analyze Ecc Attentions
-    // Analyze CEs
+    do
+    {
+        // Analyze Ecc Attentions
+        uint32_t eccAttns;
+        o_rc = checkEccFirs<T>( iv_chip, eccAttns );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "checkEccFirs(0x%08x) failed",
+                    iv_chip->getHuid() );
+            break;
+        }
 
-    o_done = true;
-    PRDF_ERR( PRDF_FUNC "function not implemented yet" );
+        o_rc = analyzeEcc( eccAttns, io_sc, o_done );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "analyzeEcc() failed." );
+            break;
+        }
+        if ( o_done ) break;
+
+        // Analyze CEs
+
+        o_done = true;
+        PRDF_ERR( PRDF_FUNC "function not implemented yet" );
+
+    }while(0);
 
     return o_rc;
 
@@ -99,6 +125,110 @@ uint32_t TpsEvent<T>::analyzeTpsPhase1_rt( STEP_CODE_DATA_STRUCT & io_sc,
 
 //------------------------------------------------------------------------------
 
+template <>
+uint32_t TpsEvent<TYPE_MCA>::analyzeEcc( const uint32_t & i_eccAttns,
+                                         STEP_CODE_DATA_STRUCT & io_sc,
+                                         bool & o_done )
+{
+    #define PRDF_FUNC "[TpsEvent<TYPE_MCA>::analyzeEcc] "
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        // If there was a UE.
+        if ( i_eccAttns & MAINT_UE )
+        {
+            PRDF_TRAC( PRDF_FUNC "UE Detected: 0x%08x,0x%02x",
+                       iv_chip->getHuid(), getKey() );
+
+            io_sc.service_data->setSignature( iv_chip->getHuid(),
+                                              PRDFSIG_MaintUE );
+
+            // At this point we don't actually have an address for the UE. The
+            // best we can do is get the address in which the command stopped.
+            MemAddr addr;
+            o_rc = getMemMaintAddr<TYPE_MCA>( iv_chip, addr );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "getMemMaintAddr(0x%08x) failed",
+                          iv_chip->getHuid() );
+                break;
+            }
+
+            o_rc = MemEcc::handleMemUe<TYPE_MCA>( iv_chip, addr,
+                                                  UE_TABLE::SCRUB_UE, io_sc );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "handleMemUe(0x%08x,0x%02x) failed",
+                          iv_chip->getHuid(), getKey() );
+                break;
+            }
+
+            // Abort this procedure because additional repairs will likely
+            // not help (also avoids complication of having UE and MPE at
+            // the same time).
+            o_done = true; break;
+        }
+
+        // If there was an IUE (MNFG only).
+        if ( mfgMode() && (i_eccAttns & MAINT_IUE) )
+        {
+            io_sc.service_data->setSignature( iv_chip->getHuid(),
+                                              PRDFSIG_MaintIUE );
+
+            o_rc = MemEcc::handleMemIue<TYPE_MCA, McaDataBundle *>( iv_chip,
+                                                                    iv_rank,
+                                                                    io_sc );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "handleMemIue(0x%08x,0x%02x) failed",
+                          iv_chip->getHuid(), getKey() );
+                break;
+            }
+
+            // If service call is set, then IUE threshold was reached.
+            if ( io_sc.service_data->queryServiceCall() )
+            {
+                PRDF_TRAC( PRDF_FUNC "IUE threshold detected: 0x%08x,0x%02x",
+                           iv_chip->getHuid(), getKey() );
+
+                // Abort this procedure because port failure will be triggered
+                // after analysis is complete.
+                o_done = true; break;
+            }
+        }
+
+        // If there was an MPE.
+        if ( i_eccAttns & MAINT_MPE )
+        {
+            io_sc.service_data->setSignature( iv_chip->getHuid(),
+                                              PRDFSIG_MaintMPE );
+
+            o_rc = MemEcc::handleMpe<TYPE_MCA, McaDataBundle *>( iv_chip,
+                iv_rank, io_sc );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "handleMpe<T>(0x%08x, 0x%02x) failed",
+                          iv_chip->getHuid(), iv_rank.getKey() );
+                break;
+            }
+
+            // Abort this procedure because the chip mark may have fixed the
+            // symbol that triggered TPS
+            o_done = true; break;
+        }
+
+    }while(0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+
+}
+
+//------------------------------------------------------------------------------
+
 // TODO: RTC 171914 Actual implementation of this procedure will be done later.
 template<>
 uint32_t TpsEvent<TYPE_MCA>::nextStep( STEP_CODE_DATA_STRUCT & io_sc,