PRD: make MNFG IPL hard CEs predictive in TPS phase 2

Change-Id: I4d950b0c08e18a5d5c5b02ed6fabfb3951f07325 Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/42078 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: Caleb N. Palmer <cnpalmer@us.ibm.com> Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com> Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/42211 Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com> Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
open-power · Jun 22, 2017 · d1924a3 · d1924a3
1 parent 676eae6
commit d1924a3
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 11 deletions.
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
@@ -210,18 +210,25 @@ uint32_t __checkEcc( ExtensibleChip * i_chip, TdQueue & io_queue,
         {
             io_sc.service_data->AddSignatureList( trgt, PRDFSIG_MaintHARD_CTE );
 
-            // Query the per-symbol counters for hard CE symbols and make
-            // callouts for each failing DIMM. Note that this function creates
-            // new error logs for each DIMM with a hard CE.
-            D db = static_cast<D>(i_chip->getDataBundle());
-            o_rc = db->getIplCeStats()->calloutHardCes(rank);
+            // Query the per-symbol counters for the hard CE symbol.
+            MemUtils::MaintSymbols symData; MemSymbol junk;
+            o_rc = MemUtils::collectCeStats<T>( i_chip, rank, symData, junk );
             if ( SUCCESS != o_rc )
             {
-                PRDF_ERR( PRDF_FUNC "calloutHardCes(0x%02x) failed",
-                          rank.getKey() );
+                PRDF_ERR( PRDF_FUNC "MemUtils::collectCeStats(0x%08x,m%ds%d) "
+                          "failed", i_chip->GetId(), rank.getMaster(),
+                          rank.getSlave() );
                 break;
             }
 
+            // The command will have finished at the end of the rank so there
+            // may be more than one symbol. Add all to the callout list.
+            for ( auto & s : symData )
+            {
+                MemoryMru memmru ( trgt, rank, s.symbol );
+                io_sc.service_data->SetCallout( memmru );
+            }
+
             // Add a TPS procedure to the queue.
             TdEntry * e = new TpsEvent<T>{ i_chip, rank };
             io_queue.push( e );

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTps_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemTps_ipl.C
@@ -82,13 +82,16 @@ uint32_t TpsEvent<TYPE_MCA>::nextStep( STEP_CODE_DATA_STRUCT & io_sc,
         //phase 1/2
         else
         {
-            //collect the CE statistics for later analysis use
+            // PHASE_1: Collect soft/intermittent CE for later analysis use.
+            // PHASE_2: Callout all hard CEs.
             McaDataBundle * db = getMcaDataBundle( iv_chip );
-            o_rc = db->getIplCeStats()->collectStats( iv_rank );
+            o_rc = ( TD_PHASE_1 == iv_phase )
+                                ? db->getIplCeStats()->collectStats(  iv_rank)
+                                : db->getIplCeStats()->calloutHardCes(iv_rank);
             if ( SUCCESS != o_rc )
             {
-                PRDF_ERR( PRDF_FUNC "collectStats(m%ds%d) failed on 0x%08x",
-                          iv_rank.getMaster(), iv_rank.getSlave(),
+                PRDF_ERR( PRDF_FUNC "collectStats/calloutHardCes(0x%02x) "
+                          "failed on 0x%08x", iv_rank.getKey(),
                           iv_chip->getHuid() );
                 break;
             }