PRD: Row Repair VCM Updates

Change-Id: I809cdc7fc2a010c5cb421bf5410d43f9bb690fee
RTC: 196073
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/63375
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://rchgit01.rchland.ibm.com/gerrit1/64982
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>

1 files changed, 198 insertions, 16 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm.C
index e0feeb362..5009b6aa6 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemVcm.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemVcm.C
@@ -112,10 +112,169 @@ uint32_t VcmEvent<TYPE_MBA>::rowRepair( STEP_CODE_DATA_STRUCT & io_sc,
 {
     #define PRDF_FUNC "[VcmEvent::rowRepair] "
 
+    PRDF_ASSERT( iv_rowRepairEnabled )
+
     uint32_t o_rc = SUCCESS;
 
     do
     {
+        // get port select
+        uint8_t l_ps = iv_mark.getSymbol().getPortSlct();
+
+        // get if the spares are available
+        bool l_spAvail, l_eccAvail;
+        o_rc = PlatServices::isSpareAvailable<TYPE_MBA>( iv_chip->getTrgt(),
+            iv_rank, l_ps, l_spAvail, l_eccAvail );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "isChipMarkOnSpare(0x%08x) failed",
+                      iv_chip->getHuid() );
+            break;
+        }
+
+        // get dimm
+        TARGETING::TargetHandle_t l_dimm =
+            PlatServices::getConnectedDimm( iv_chip->getTrgt(), iv_rank,
+                                            l_ps );
+
+        // If scrub stops on first MCE, and static row repair
+        // not supported or both spare and chip mark used
+        if ( 1 == iv_mceCount && ( !l_spAvail && !l_eccAvail ) )
+        {
+            // Record bad DQs in VPD - done when verified()
+            // No need to continue scrubbing, VCM verified, VCM done.
+            o_done = true;
+        }
+        // Else if scrub stops on first MCE and static row repair
+        // supported
+        else if ( 1 == iv_mceCount )
+        {
+            MemRowRepair l_rowRepair;
+            o_rc = getRowRepairData<TYPE_MBA>( l_dimm, iv_rank, l_rowRepair );
+            if ( SUCCESS != o_rc )
+            {
+                PRDF_ERR( PRDF_FUNC "getRowRepairData(0x%08x, 0x%02x)",
+                          PlatServices::getHuid(l_dimm), iv_rank.getKey() );
+                break;
+            }
+
+            // If the port, dimm, master rank has previous row repair in VPD
+            if ( l_rowRepair.isValid() )
+            {
+                // If previous repair for same DRAM
+                if ( l_rowRepair.getRowRepairDram() ==
+                     iv_mark.getSymbol().getDram() )
+                {
+                    // Clear previous row repair from VPD
+                    o_rc = clearRowRepairData<TYPE_MBA>( l_dimm, iv_rank );
+                    if ( SUCCESS != o_rc )
+                    {
+                        PRDF_ERR( PRDF_FUNC "clearRowRepairData"
+                                  "(0x%08x, 0x%02x) failed",
+                                  PlatServices::getHuid(l_dimm),
+                                  iv_rank.getKey() );
+                        break;
+                    }
+
+                    // Record bad DQs in VPD - done when verified()
+                    // Signature: "VCM: verified: previous PPR on same DRAM"
+                    io_sc.service_data->setSignature( iv_chip->getHuid(),
+                        PRDFSIG_VcmVerSameDram );
+
+                    // No need to continue scrubbing, VCM verified, VCM done
+                    o_done = true;
+                }
+                // Else if previous repair for different DRAM
+                else
+                {
+                    // Leave previous row repair in VPD
+                    // Record bad DQs in VPD - done when verified()
+                    // Signature:"VCM: verified: previous PPR on
+                    // different DRAM"
+                    io_sc.service_data->setSignature( iv_chip->getHuid(),
+                        PRDFSIG_VcmVerDiffDram );
+
+                    // No need to continue scrubbing, VCM verified, VCM done
+                    o_done = true;
+                }
+            }
+            // Else if no previous row repair
+            else
+            {
+                // Signature: "VCM: verified: first MCE"
+                io_sc.service_data->setSignature( iv_chip->getHuid(),
+                    PRDFSIG_VcmVerFirstMce );
+
+                // Record bad DQs in VPD - done when verified()
+                // Remember address
+                MemAddr l_addr;
+                o_rc = getMemMaintAddr<TYPE_MBA>( iv_chip,
+                                                  iv_rowRepairFailAddr );
+                if ( SUCCESS != o_rc )
+                {
+                    PRDF_ERR( PRDF_FUNC "getMemMaintAddr(0x%08x) failed",
+                              iv_chip->getHuid() );
+                    break;
+                }
+
+                // Continue scrub, don't set procedure to done
+            }
+        }
+        // Else if scrub stops on second MCE
+        else if ( iv_mceCount > 1 )
+        {
+            // Since at least 2 bad rows, don't bother with row repair
+            // No need to continue scrubbing, VCM verified, VCM done
+            o_done = true;
+
+            // Signature: "VCM: verified: second MCE"
+            io_sc.service_data->setSignature( iv_chip->getHuid(),
+                PRDFSIG_VcmVerSecMce );
+        }
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+template<>
+uint32_t VcmEvent<TYPE_MBA>::rowRepairEndRank( STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[VcmEvent::rowRepairEndRank] "
+
+    PRDF_ASSERT( !iv_canResumeScrub );
+    PRDF_ASSERT( iv_rowRepairEnabled );
+    PRDF_ASSERT( 0 != iv_mceCount );
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        // get dimm
+        uint8_t l_ps = iv_mark.getSymbol().getPortSlct();
+        TARGETING::TargetHandle_t l_dimm =
+            PlatServices::getConnectedDimm( iv_chip->getTrgt(), iv_rank,
+                                            l_ps );
+
+        // If scrub gets to the end of the master rank with an MCE
+        // Update VPD with row repair
+        o_rc = setRowRepairData<TYPE_MBA>( l_dimm, iv_rank,
+            iv_rowRepairFailAddr, iv_mark.getSymbol().getDram() );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "setRowRepairData(0x%08x, 0x%02x) "
+                      "failed", PlatServices::getHuid(l_dimm),
+                      iv_rank.getKey() );
+            break;
+        }
+
+        // Signature: "VCM: verified: common row fail"
+        io_sc.service_data->setSignature( iv_chip->getHuid(),
+                                          PRDFSIG_VcmVerRowFail );
+
+        // VCM verified, VCM done
 
     } while (0);
 
@@ -196,41 +355,64 @@ uint32_t VcmEvent<TYPE_MBA>::handlePhaseComplete( const uint32_t & i_eccAttns,
         {
             if ( i_eccAttns & MAINT_MCE )
             {
-                if ( iv_rowRepairEnabled )
+                iv_mceCount++;
+
+                // Only need to call verified on the first mce we hit
+                if ( 1 == iv_mceCount )
                 {
-                    o_rc = rowRepair( io_sc, o_done );
+                    o_rc = verified( io_sc );
                     if ( SUCCESS != o_rc )
                     {
-                        PRDF_ERR( PRDF_FUNC "rowRepair() failed on 0x%08x",
+                        PRDF_ERR( PRDF_FUNC "verified() failed on 0x%08x",
                                   iv_chip->getHuid() );
                         break;
                     }
                 }
-                else
+
+                if ( iv_rowRepairEnabled )
                 {
-                    o_rc = verified( io_sc );
+                    o_rc = rowRepair( io_sc, o_done );
                     if ( SUCCESS != o_rc )
                     {
-                        PRDF_ERR( PRDF_FUNC "verified() failed on 0x%08x",
+                        PRDF_ERR( PRDF_FUNC "rowRepair() failed on 0x%08x",
                                   iv_chip->getHuid() );
                         break;
                     }
-
+                    if ( o_done ) break;
+                }
+                else
+                {
                     o_done = true; // Procedure is complete.
+                    break;
                 }
             }
-            else if ( !iv_canResumeScrub )
+
+            if ( !iv_canResumeScrub )
             {
-                // The chip mark is not verified and the command has reached the
-                // end of the rank. So this is a false alarm.
-                o_rc = falseAlarm( io_sc );
-                if ( SUCCESS != o_rc )
+                // If row repair is enabled, we reached the end of the rank, and
+                // we got an MCE, we need to apply the row repair.
+                if ( iv_rowRepairEnabled && 0 != iv_mceCount )
                 {
-                    PRDF_ERR( PRDF_FUNC "falseAlarm() failed on 0x%08x",
-                              iv_chip->getHuid() );
-                    break;
+                    o_rc = rowRepairEndRank( io_sc );
+                    if ( SUCCESS != o_rc )
+                    {
+                        PRDF_ERR( PRDF_FUNC "rowRepairEndRank() failed on "
+                                  "0x%08x", iv_chip->getHuid() );
+                        break;
+                    }
+                }
+                else
+                {
+                    // The chip mark is not verified and the command has reached
+                    // the end of the rank. So this is a false alarm.
+                    o_rc = falseAlarm( io_sc );
+                    if ( SUCCESS != o_rc )
+                    {
+                        PRDF_ERR( PRDF_FUNC "falseAlarm() failed on 0x%08x",
+                                  iv_chip->getHuid() );
+                        break;
+                    }
                 }
-
                 o_done = true; // Procedure is complete.
             }
         }