PRD: Update handleRrFo

Change-Id: I86bbd32ff782aa30287f493c72142988b4618922
RTC: 171874
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/48652
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@us.ibm.com>
Tested-by: Jenkins Server <pfd-jenkins+hostboot@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/gerrit1/49483
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>

5 files changed, 290 insertions, 3 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H
index 33f3187ea..e9cdfd051 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H
@@ -108,6 +108,24 @@ class MemTdCtlr
      */
     void pushToQueue( TdEntry * i_entry ) { iv_queue.push(i_entry); }
 
+    #ifdef __HOSTBOOT_RUNTIME
+
+    /**
+     * @brief Handles reset-reload or FO scenario.
+     * @note  This function will check if PRD was unable to restart maintenance
+     *        command before R/R or FO. In that scenario, this function will
+     *        start maintenance command. As during R/R or F/O we do not have any
+     *        mechanism to restore the complete state of TD controller, we will
+     *        not start any interrupted or pending TD procedure. We will only
+     *        start BG scrub. If we found any chip marks during TD state
+     *        machine initialize we will start VCM procedure rather than
+     *        BG scrub.
+     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+     */
+    uint32_t handleRrFo();
+
+    #endif
+
   private:
 
     /**
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
index c6215ba30..d65465964 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
@@ -914,6 +914,230 @@ uint32_t MemTdCtlr<TYPE_MBA>::unmaskEccAttns()
 
 //------------------------------------------------------------------------------
 
+template<>
+uint32_t MemTdCtlr<TYPE_MCBIST>::handleRrFo()
+{
+    #define PRDF_FUNC "[MemTdCtlr<TYPE_MCBIST>::handleRrFo] "
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        // Check if maintenance command complete attention is set.
+        SCAN_COMM_REGISTER_CLASS * mcbistfir =
+                                iv_chip->getRegister("MCBISTFIR");
+        o_rc = mcbistfir->Read();
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on MCBISTFIR");
+            break;
+        }
+
+        // If there is a command complete attention, nothing to do, break out.
+        if ( mcbistfir->IsBitSet(10) ||  mcbistfir->IsBitSet(12) )
+            break;
+
+
+        // Check if a command is not running.
+        // If bit 0 of MCB_CNTLSTAT is on, a mcbist run is in progress.
+        SCAN_COMM_REGISTER_CLASS * mcb_cntlstat =
+            iv_chip->getRegister("MCB_CNTLSTAT");
+        o_rc = mcb_cntlstat->Read();
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on MCB_CNTLSTAT" );
+            break;
+        }
+
+        // If a command is not running, set command complete attn, break.
+        if ( !mcb_cntlstat->IsBitSet(0) )
+        {
+            SCAN_COMM_REGISTER_CLASS * mcbistfir_or =
+                iv_chip->getRegister("MCBISTFIR_OR");
+            mcbistfir_or->SetBit( 10 );
+
+            mcbistfir_or->Write();
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "Write() failed on MCBISTFIR_OR" );
+            }
+            break;
+        }
+
+        // Check if there are unverified chip marks.
+        std::vector<TdRankListEntry> vectorList = iv_rankList.getList();
+
+        for ( auto & entry : vectorList )
+        {
+            ExtensibleChip * mcaChip = entry.getChip();
+            MemRank rank = entry.getRank();
+
+            // Get the chip mark
+            MemMark chipMark;
+            o_rc = MarkStore::readChipMark<TYPE_MCA>( mcaChip, rank, chipMark );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "readChipMark<TYPE_MCA>(0x%08x,%d) "
+                        "failed", mcaChip->getHuid(), rank.getMaster() );
+                break;
+            }
+
+            if ( !chipMark.isValid() ) continue; // no chip mark present
+
+            // Get the DQ Bitmap data.
+            TargetHandle_t mcaTrgt = mcaChip->GetChipHandle();
+            MemDqBitmap<DIMMS_PER_RANK::MCA> dqBitmap;
+
+            o_rc = getBadDqBitmap<DIMMS_PER_RANK::MCA>(mcaTrgt, rank, dqBitmap);
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "getBadDqBitmap<DIMMS_PER_RANK::MCA>"
+                        "(0x%08x, %d)", getHuid(mcaTrgt), rank.getMaster() );
+                break;
+            }
+
+            // Check if the chip mark is verified or not.
+            bool cmVerified = false;
+            o_rc = dqBitmap.isChipMark( chipMark.getSymbol(), cmVerified );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "dqBitmap.isChipMark failed." );
+                break;
+            }
+
+            // If there are any unverified chip marks, stop the command, break.
+            if ( !cmVerified )
+            {
+                o_rc = stopBgScrub<TYPE_MCBIST>( iv_chip );
+                if ( SUCCESS != o_rc )
+                {
+                    PRDF_ERR( PRDF_FUNC "stopBgScrub<TYPE_MCBIST>(0x%08x) "
+                              "failed", iv_chip->getHuid() );
+                }
+                break;
+            }
+        }
+
+    } while (0);
+
+    return o_rc;
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
+template<>
+uint32_t MemTdCtlr<TYPE_MBA>::handleRrFo()
+{
+    #define PRDF_FUNC "[MemTdCtlr<TYPE_MBA>::handleRrFo] "
+
+     uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        // Check if maintenance command complete attention is set.
+        SCAN_COMM_REGISTER_CLASS * mbaspa =
+                                iv_chip->getRegister("MBASPA");
+        o_rc = mbaspa->Read();
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on MBASPA");
+            break;
+        }
+
+        // If there is a command complete attention, nothing to do, break out.
+        if ( mbaspa->IsBitSet(0) ||  mbaspa->IsBitSet(8) )
+            break;
+
+        // Check if a maintenance command is running currently.
+        SCAN_COMM_REGISTER_CLASS * mbmsr =
+                                iv_chip->getRegister("MBMSR");
+
+        o_rc = mbmsr->Read();
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on MBMSR");
+            break;
+        }
+
+        // If a command is not running, set command complete attn, break.
+        if ( !mbmsr->IsBitSet(0) )
+        {
+            SCAN_COMM_REGISTER_CLASS * mbaspa_or =
+                iv_chip->getRegister("MBASPA_OR");
+            mbaspa_or->SetBit( 0 );
+
+            mbaspa_or->Write();
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "Write() failed on MBASPA_OR" );
+            }
+            break;
+        }
+
+        // Check if there are unverified chip marks.
+        std::vector<TdRankListEntry> vectorList = iv_rankList.getList();
+
+        for ( auto & entry : vectorList )
+        {
+            ExtensibleChip * mbaChip = entry.getChip();
+            MemRank rank = entry.getRank();
+
+            // Get the chip mark
+            MemMark chipMark;
+            o_rc = MarkStore::readChipMark<TYPE_MBA>( mbaChip, rank, chipMark );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "readChipMark<TYPE_MBA>(0x%08x,%d) "
+                        "failed", mbaChip->getHuid(), rank.getMaster() );
+                break;
+            }
+
+            if ( !chipMark.isValid() ) continue; // no chip mark present
+
+            // Get the DQ Bitmap data.
+            TargetHandle_t mbaTrgt = mbaChip->GetChipHandle();
+            MemDqBitmap<DIMMS_PER_RANK::MBA> dqBitmap;
+
+            o_rc = getBadDqBitmap<DIMMS_PER_RANK::MBA>(mbaTrgt, rank, dqBitmap);
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "getBadDqBitmap<DIMMS_PER_RANK::MBA>"
+                        "(0x%08x, %d)", getHuid(mbaTrgt), rank.getMaster() );
+                break;
+            }
+
+            // Check if the chip mark is verified or not.
+            bool cmVerified = false;
+            o_rc = dqBitmap.isChipMark( chipMark.getSymbol(), cmVerified );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "dqBitmap.isChipMark failed." );
+                break;
+            }
+
+            // If there are any unverified chip marks, stop the command, break.
+            if ( !cmVerified )
+            {
+                o_rc = stopBgScrub<TYPE_MBA>( iv_chip );
+                if ( SUCCESS != o_rc )
+                {
+                    PRDF_ERR( PRDF_FUNC "stopBgScrub<TYPE_MBA>(0x%08x) failed",
+                            iv_chip->getHuid() );
+                }
+                break;
+            }
+        }
+
+    } while (0);
+
+    return o_rc;
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
+
 // Avoid linker errors with the template.
 template class MemTdCtlr<TYPE_MCBIST>;
 template class MemTdCtlr<TYPE_MBA>;
diff --git a/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.C b/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.C
index 241d835f2..88c72fdf3 100644
--- a/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.C
+++ b/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -35,6 +35,7 @@
 #include <prdfPlatServices.H>
 #include <prdfTrace.H>
 #include <prdfMemBgScrub.H>
+#include <prdfP9McbistDataBundle.H>
 
 using namespace TARGETING;
 
@@ -43,6 +44,7 @@ namespace PRDF
 
 using namespace PlatServices;
 
+#ifndef __HOSTBOOT_RUNTIME
 int32_t McbistDomain::startScrub()
 {
     #define PRDF_FUNC "[McbistDomain::startScrub] "
@@ -72,5 +74,38 @@ int32_t McbistDomain::startScrub()
 
     #undef PRDF_FUNC
 }
+#endif
+
+#ifdef __HOSTBOOT_RUNTIME
+void McbistDomain::handleRrFo()
+{
+    #define PRDF_FUNC "[McbistDomain::handleRrFo] "
+
+    do
+    {
+        uint32_t domainSize = GetSize();
+        // Iterate all MCBISTs in the domain.
+        for ( uint32_t i = 0; i < domainSize; ++i )
+        {
+            RuleChip * mcbChip = LookUp(i);
+
+            // Start background scrub if required.
+            McbistDataBundle * mcbdb = getMcbistDataBundle( mcbChip );
+            int32_t l_rc = mcbdb->getTdCtlr()->handleRrFo();
+            if ( SUCCESS != l_rc )
+            {
+                // Let us not fail here. If problem is contained within a MCBIST
+                // we will discover it again during normal TD procedures.
+                PRDF_ERR( PRDF_FUNC "handleRrFo() failed: MCBIST=0x%08x",
+                          mcbChip->GetId() );
+                continue; // Keep going.
+            }
+        }
+
+    } while (0);
+
+    #undef PRDF_FUNC
+}
+#endif
 
 } // end namespace PRDF
diff --git a/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.H b/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.H
index 702c97af9..1e9b15d7e 100755
--- a/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.H
+++ b/src/usr/diag/prdf/plat/mem/prdfP9McbistDomain.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -54,6 +54,16 @@ class McbistDomain : public RuleChipDomain
     virtual bool Query( ATTENTION_TYPE i_attnType )
     {  return false;  }
 
+    #ifdef __HOSTBOOT_RUNTIME
+
+    /**
+     * @brief Starts memory background scrubbing or VCM procedure for MCBIST
+     *        during R/R and F/O if required.
+     */
+    void handleRrFo();
+
+    #endif
+
     #ifndef __HOSTBOOT_RUNTIME
 
     /**
diff --git a/src/usr/diag/prdf/plat/mem/prdf_plat_mem_hb_only.mk b/src/usr/diag/prdf/plat/mem/prdf_plat_mem_hb_only.mk
index a89ec90f6..1c7770ee4 100644
--- a/src/usr/diag/prdf/plat/mem/prdf_plat_mem_hb_only.mk
+++ b/src/usr/diag/prdf/plat/mem/prdf_plat_mem_hb_only.mk
@@ -40,6 +40,7 @@ prd_incpath += ${PRD_SRC_PATH}/plat/mem
 # plat/mem/ (non-rule plugin related)
 prd_obj += prdfMemScrubUtils.o
 prd_obj += prdfMemTdCtlr.o
+prd_obj += prdfP9McbistDomain.o
 
 # plat/mem/ (rule plugin related)
 prd_rule_plugin += prdfCenMba.o
@@ -56,7 +57,6 @@ ifneq (${HOSTBOOT_RUNTIME},1)
 prd_obj += prdfMemTdCtlr_ipl.o
 prd_obj += prdfMemTps_ipl.o
 prd_obj += prdfMemVcm_ipl.o
-prd_obj += prdfP9McbistDomain.o
 prd_obj += prdfMemIplCeStats.o
 prd_obj += prdfRestoreDramRepairs.o