Enable mss_scrub on MPIPL if NVDIMM is configured

NVDIMM restore blows away the scrub settings so it needs to be
reinitialized during MPIPL

Change-Id: I67f74c71e98b8907351eb7477bf197a941aefb8d
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/71786
Tested-by: Jenkins Server <pfd-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>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Reviewed-by: Caleb N. Palmer <cnpalmer@us.ibm.com>
Reviewed-by: TSUNG K. YEUNG <tyeung@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>

6 files changed, 122 insertions, 80 deletions

diff --git a/src/include/usr/diag/prdf/prdfMain_ipl.H b/src/include/usr/diag/prdf/prdfMain_ipl.H
index 933a3d47f..08b626ce2 100644
--- a/src/include/usr/diag/prdf/prdfMain_ipl.H
+++ b/src/include/usr/diag/prdf/prdfMain_ipl.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2014,2017                        */
+/* Contributors Listed Below - COPYRIGHT 2014,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -65,16 +65,13 @@ extern int32_t analyzeIplCEStats( const TARGETING::TargetHandle_t i_mba,
                                   bool &o_calloutMade );
 
 /**
- * @brief  Starts memory background scrubbing on all MBAs in the node.
- *
- * This will only start the initial fast scrub of all memory. PRD on the FSP or
- * HostServices (SP-less machines) will need to resume normal background
- * scrubbing once this initial fast scrub is complete.
- *
+ * @brief  Starts memory background scrubbing on the given target.
+ * @param  i_trgt A target that contains the maintenance command logic (i.e.
+ *                MCBIST or MBA).
  * @return If an error log is returned, then some internal function failed. See
  *         the FFDC in the error log for failure details.
  */
-extern errlHndl_t startScrub();
+extern errlHndl_t startScrub( const TARGETING::TargetHandle_t i_trgt );
 
 #ifdef CONFIG_ENABLE_CHECKSTOP_ANALYSIS
 
diff --git a/src/include/usr/isteps/istep16list.H b/src/include/usr/isteps/istep16list.H
index a6a5f7df7..37cc26ed4 100644
--- a/src/include/usr/isteps/istep16list.H
+++ b/src/include/usr/isteps/istep16list.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2012,2018                        */
+/* Contributors Listed Below - COPYRIGHT 2012,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -141,7 +141,11 @@ namespace   INITSERVICE
                 // Hostboot PRD should no longer be called after this step.
                 ISTEPNAME(16,04,"mss_scrub"),
                 ISTEP_16::call_mss_scrub,
+#ifdef CONFIG_NVDIMM
+                { START_FN, EXT_IMAGE, NORMAL_IPL_OP | MPIPL_OP, false }
+#else
                 { START_FN, EXT_IMAGE, NORMAL_IPL_OP, false }
+#endif
         },
         {
                 ISTEPNAME(16,05,"host_load_io_ppe"),
diff --git a/src/include/usr/isteps/istep_reasoncodes.H b/src/include/usr/isteps/istep_reasoncodes.H
index 4aa6a78a0..09b12abdc 100644
--- a/src/include/usr/isteps/istep_reasoncodes.H
+++ b/src/include/usr/isteps/istep_reasoncodes.H
@@ -65,6 +65,7 @@ namespace ISTEP
         MOD_SMP_WRAP_PROC_IOVALID                         = 0x22,
         MOD_APPLY_HCODE_GEN_CPU_REGS                      = 0x23,
         MOD_GET_LANEMASK_FROM_HX_KEYWORD                  = 0x24,
+        MOD_MSS_SCRUB                                     = 0x25,
     };
 
     /**
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemBgScrub.H b/src/usr/diag/prdf/plat/mem/prdfMemBgScrub.H
index c04d9dcbb..c937e423b 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemBgScrub.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemBgScrub.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016,2018                        */
+/* Contributors Listed Below - COPYRIGHT 2016,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
diff --git a/src/usr/diag/prdf/prdfMain_ipl.C b/src/usr/diag/prdf/prdfMain_ipl.C
index 8241d925f..b6959e00f 100644
--- a/src/usr/diag/prdf/prdfMain_ipl.C
+++ b/src/usr/diag/prdf/prdfMain_ipl.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2014,2018                        */
+/* Contributors Listed Below - COPYRIGHT 2014,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -40,10 +40,9 @@
 
 // Platform includes
 #include <prdfCenMbaDataBundle.H>
-#include <prdfCenMbaDomain.H>
 #include <prdfPlatServices.H>
 #include <prdfP9McaDataBundle.H>
-#include <prdfP9McbistDomain.H>
+#include <prdfMemBgScrub.H>
 
 // Custom compile configs
 #include <config.h>
@@ -115,7 +114,7 @@ int32_t analyzeIplCEStats( TargetHandle_t i_trgt, bool &o_calloutMade )
 
 //------------------------------------------------------------------------------
 
-errlHndl_t startScrub()
+errlHndl_t startScrub( const TargetHandle_t i_trgt )
 {
     #define PRDF_FUNC "[PRDF::startScrub] "
     PRDF_ENTER( PRDF_FUNC );
@@ -123,63 +122,40 @@ errlHndl_t startScrub()
     errlHndl_t o_errl = nullptr;
 
     int32_t l_rc = SUCCESS;
-    HUID nodeId = INVALID_HUID;
 
     // will unlock when going out of scope
     PRDF_SYSTEM_SCOPELOCK;
 
     do
     {
-        // Since the last refresh is in istep10 host_prd_hwreconfig,
-        // it may be good to call it again here at istep16 mss_scrub
-        // to remove any non-functional MBAs from PRD system model.
-        o_errl = noLock_refresh();
-        // This shouldn't return any error but if it does, break out
-        if( nullptr != o_errl )
+        // Get the PRD chip object.
+        ExtensibleChip * chip = (ExtensibleChip *)systemPtr->GetChip(i_trgt);
+        if ( nullptr == chip )
         {
-            PRDF_ERR( PRDF_FUNC "noLock_refresh() failed" );
-            break;
-        }
-
-        // This is run in Hostboot so there should only be one node.
-        TargetHandleList list = getFunctionalTargetList( TYPE_NODE );
-        if ( 1 != list.size() )
-        {
-            PRDF_ERR( PRDF_FUNC "getFunctionalTargetList(TYPE_NODE) failed" );
+            PRDF_ERR( PRDF_FUNC "unable to find chip object for given target: "
+                                "0x%08x", getHuid(i_trgt) );
             l_rc = FAIL; break;
         }
-        nodeId = getHuid(list[0]);
 
-        PRDF_ENTER( PRDF_FUNC "HUID=0x%08x", nodeId );
-
-        // Start background scrubbing.
-        TARGETING::MODEL procModel = getChipModel( getMasterProc() );
-        if ( MODEL_CUMULUS == procModel )
+        // Start background scrubbing on this target.
+        switch ( chip->getType() )
         {
-            MbaDomain * domain = (MbaDomain *)systemPtr->GetDomain(MBA_DOMAIN);
-            l_rc = domain->startScrub();
+            case TYPE_MBA:    startInitialBgScrub<TYPE_MBA>(   chip); break;
+            case TYPE_MCBIST: startInitialBgScrub<TYPE_MCBIST>(chip); break;
+            default:
+                PRDF_ERR( PRDF_FUNC "Unsupported maintenance target type "
+                          "0x%02x", chip->getType() );
+                l_rc = FAIL;
         }
-        else if ( MODEL_NIMBUS == procModel )
-        {
-            McbistDomain * domain =
-                            (McbistDomain *)systemPtr->GetDomain(MCBIST_DOMAIN);
-            l_rc = domain->startScrub();
-        }
-        else
-        {
-            PRDF_ERR(PRDF_FUNC "Master PROC model %d not supported", procModel);
-            PRDF_ASSERT(false);
-        }
-
-        PRDF_EXIT( PRDF_FUNC "HUID=0x%08x", nodeId );
+        if ( SUCCESS != l_rc ) break;
 
     } while (0);
 
-    if (( SUCCESS != l_rc ) && (NULL == o_errl))
+    if ( SUCCESS != l_rc )
     {
         // Get user data
-        uint64_t ud12 = PRDF_GET_UINT64_FROM_UINT32( nodeId, __LINE__ );
-        uint64_t ud34 = PRDF_GET_UINT64_FROM_UINT32( 0,      0        );
+        uint64_t ud12 = PRDF_GET_UINT64_FROM_UINT32( getHuid(i_trgt), 0 );
+        uint64_t ud34 = PRDF_GET_UINT64_FROM_UINT32( 0,               0 );
 
         // Create error log
         o_errl = new ERRORLOG::ErrlEntry(
@@ -193,7 +169,9 @@ errlHndl_t startScrub()
                                      HWAS::SRCI_PRIORITY_HIGH );
 
         // Add traces
-        o_errl->collectTrace( PRDF_COMP_NAME, 512 );
+        o_errl->collectTrace( PRDF_COMP_NAME,      512 );
+        o_errl->collectTrace( FAPI_TRACE_NAME,     256 );
+        o_errl->collectTrace( FAPI_IMP_TRACE_NAME, 256 );
     }
 
     PRDF_EXIT( PRDF_FUNC );
diff --git a/src/usr/isteps/istep16/call_mss_scrub.C b/src/usr/isteps/istep16/call_mss_scrub.C
index 55e186334..aa8fd3861 100644
--- a/src/usr/isteps/istep16/call_mss_scrub.C
+++ b/src/usr/isteps/istep16/call_mss_scrub.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2015,2017                        */
+/* Contributors Listed Below - COPYRIGHT 2015,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -44,9 +44,11 @@ namespace ISTEP_16
 {
 void* call_mss_scrub (void *io_pArgs)
 {
+    #define ISTEP_FUNC "call_mss_scrub: "
+
     IStepError l_stepError;
 
-    TRACDCOMP( ISTEPS_TRACE::g_trac_isteps_trace, "call_mss_scrub entry" );
+    TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, ISTEP_FUNC "entry" );
 
     errlHndl_t errl = nullptr;
 
@@ -56,57 +58,117 @@ void* call_mss_scrub (void *io_pArgs)
         {
             // There are performance issues and some functional deficiencies
             // that make background scrub problematic in SIMICs.
-            TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, "Skipping background "
-                       "scrub in SIMICs" );
+            TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, ISTEP_FUNC
+                       "Background scrubbing not supported in SIMICs" );
+            break;
         }
-        else
+
+        TargetHandle_t sysTrgt = nullptr;
+        targetService().getTopLevelTarget( sysTrgt );
+
+        TargetHandle_t masterProc = nullptr;
+        targetService().masterProcChipTargetHandle( masterProc );
+
+        // Determine which target type runs the maintenance commands.
+        TARGETING::MODEL masterProcModel = masterProc->getAttr<ATTR_MODEL>();
+        TARGETING::TYPE maintTrgtType;
+        switch ( masterProcModel )
         {
-            // Start background scrubbing.
-            errl = PRDF::startScrub();
+            case MODEL_CUMULUS: maintTrgtType = TYPE_MBA;       break;
+            case MODEL_NIMBUS:  maintTrgtType = TYPE_MCBIST;    break;
+            default:
+                TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, ISTEP_FUNC
+                           "Master PROC model %d not supported",
+                           masterProcModel );
+                /*@
+                * @errortype
+                * @moduleid     ISTEP::MOD_MSS_SCRUB
+                * @reasoncode   ISTEP::RC_INVALID_TARGET_TYPE
+                * @userdata1    Master processor model
+                * @userdata2    unused
+                * @devdesc      The master processor model is unsupported.
+                * @custdesc     A problem occurred during the IPL of the system.
+                */
+                errl = new ERRORLOG::ErrlEntry(
+                                    ERRORLOG::ERRL_SEV_UNRECOVERABLE,
+                                    ISTEP::MOD_MSS_SCRUB,
+                                    ISTEP::RC_INVALID_TARGET_TYPE,
+                                    masterProcModel, 0 );
+        }
+        if ( nullptr != errl ) break;
+
+        // Start background scrubbing on all targets of this maintenance type.
+        TargetHandleList maintList; getAllChiplets( maintList, maintTrgtType );
+        for ( auto & maintTrgt : maintList )
+        {
+            bool start = true; // initially true except for MP-IPL conditions.
+
+#ifdef CONFIG_NVDIMM
+            // During MP-IPLs, We only want to start background scrubbing on
+            // maintenance targets that have connected NVDIMMs.
+            if ( sysTrgt->getAttr<ATTR_IS_MPIPL_HB>() )
+            {
+                start = false; // Only true if there is an NVDIMM.
+
+                // Find at least one DIMM behind this MCBIST that is an NVDIMM.
+                TargetHandleList dimmList;
+                getChildAffinityTargets( dimmList, maintTrgt, CLASS_NA,
+                                         TYPE_DIMM );
+                for ( auto & dimmTrgt : dimmList )
+                {
+                    start = isNVDIMM(dimmTrgt);
+                    if ( start ) break;
+                }
+            }
+#endif
+            // Continue to the next target if we are unable to start background
+            // scrubbing on this target.
+            if ( !start ) continue;
+
+            // Start the command on this target.
+            errl = PRDF::startScrub( maintTrgt );
             if ( nullptr != errl )
             {
-                TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace,
-                           "PRDF::startScrub() failed" );
+                TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, ISTEP_FUNC
+                           "PRDF::startScrub(0x%08x) failed",
+                           get_huid(maintTrgt) );
                 break;
             }
-        }
 
-        // Nimbus chips require us to unmask some additional FIR bits. Note that
-        // this is not needed on Cumulus based systems because this is already
-        // contained within the other Centaur HWPs.
-        TargetHandle_t masterProc = nullptr;
-        targetService().masterProcChipTargetHandle(masterProc);
-        if ( MODEL_NIMBUS == masterProc->getAttr<ATTR_MODEL>() )
-        {
-            TargetHandleList trgtList; getAllChiplets( trgtList, TYPE_MCBIST );
-
-            for ( auto & tt : trgtList )
+            // Nimbus chips require us to unmask some additional FIR bits. Note
+            // that this is not needed on Cumulus based systems because this is
+            // already contained within the other Centaur HWPs.
+            if ( TYPE_MCBIST == maintTrgtType )
             {
-                fapi2::Target<fapi2::TARGET_TYPE_MCBIST> ft ( tt );
+                fapi2::Target<fapi2::TARGET_TYPE_MCBIST> ft ( maintTrgt );
 
                 FAPI_INVOKE_HWP( errl, mss::unmask::after_background_scrub, ft);
                 if ( nullptr != errl )
                 {
-                    TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace,
+                    TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, ISTEP_FUNC
                                "mss::unmask::after_background_scrub(0x%08x) "
-                               "failed", get_huid(tt) );
+                               "failed", get_huid(maintTrgt) );
                     break;
                 }
             }
-            if ( nullptr != errl ) break;
         }
+        if ( nullptr != errl ) break;
 
     } while (0);
 
     if ( nullptr != errl )
     {
         l_stepError.addErrorDetails( errl );
+        errl->collectTrace( "ISTEPS_TRACE", 256 );
         errlCommit( errl, HWPF_COMP_ID );
     }
 
-    TRACDCOMP( ISTEPS_TRACE::g_trac_isteps_trace, "call_mss_scrub exit" );
+    TRACFCOMP( ISTEPS_TRACE::g_trac_isteps_trace, ISTEP_FUNC "exit" );
+
     // end task, returning any errorlogs to IStepDisp
     return l_stepError.getErrorHandle();
+
+    #undef ISTEP_FUNC
 }
 
 };