PRD: added PlatServices::isRowRepairEnabled()

Change-Id: Ia2ac63ecb1b07af6dab75afa9e00fe966ba11dee RTC: 196073 Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/63987 Reviewed-by: Caleb N. Palmer <cnpalmer@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://rchgit01.rchland.ibm.com/gerrit1/64559 Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com> Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com> Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>
open-power · Aug 16, 2018 · e55c53f · e55c53f
1 parent 890ac53
commit e55c53f
Show file tree

Hide file tree

Showing 7 changed files with 129 additions and 9 deletions.
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm.C
@@ -106,6 +106,26 @@ uint32_t VcmEvent<TYPE_MCA>::handlePhaseComplete( const uint32_t & i_eccAttns,
 //
 //##############################################################################
 
+template<>
+uint32_t VcmEvent<TYPE_MBA>::rowRepair( STEP_CODE_DATA_STRUCT & io_sc,
+                                        bool & o_done )
+{
+    #define PRDF_FUNC "[VcmEvent::rowRepair] "
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
 template<>
 uint32_t VcmEvent<TYPE_MBA>::startNextPhase( STEP_CODE_DATA_STRUCT & io_sc )
 {
@@ -176,16 +196,28 @@ uint32_t VcmEvent<TYPE_MBA>::handlePhaseComplete( const uint32_t & i_eccAttns,
         {
             if ( i_eccAttns & MAINT_MCE )
             {
-                // The chip mark has been verified.
-                o_rc = verified( io_sc );
-                if ( SUCCESS != o_rc )
+                if ( iv_rowRepairEnabled )
                 {
-                    PRDF_ERR( PRDF_FUNC "verified() failed on 0x%08x",
-                              iv_chip->getHuid() );
-                    break;
+                    o_rc = rowRepair( io_sc, o_done );
+                    if ( SUCCESS != o_rc )
+                    {
+                        PRDF_ERR( PRDF_FUNC "rowRepair() failed on 0x%08x",
+                                  iv_chip->getHuid() );
+                        break;
+                    }
+                }
+                else
+                {
+                    o_rc = verified( io_sc );
+                    if ( SUCCESS != o_rc )
+                    {
+                        PRDF_ERR( PRDF_FUNC "verified() failed on 0x%08x",
+                                  iv_chip->getHuid() );
+                        break;
+                    }
+
+                    o_done = true; // Procedure is complete.
                 }
-
-                o_done = true; // Procedure is complete.
             }
             else if ( !iv_canResumeScrub )
             {

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm.H b/src/usr/diag/prdf/plat/mem/prdfMemVcm.H
@@ -55,7 +55,8 @@ class VcmEvent : public TdEntry
      */
     VcmEvent<T>( ExtensibleChip * i_chip, const MemRank & i_rank,
                  const MemMark & i_mark ) :
-        TdEntry(VCM_EVENT, i_chip, i_rank), iv_mark(i_mark)
+        TdEntry(VCM_EVENT, i_chip, i_rank), iv_mark(i_mark),
+        iv_rowRepairEnabled(PlatServices::isRowRepairEnabled<T>(i_chip,i_rank))
     {
         PRDF_ASSERT( nullptr != i_chip );
         PRDF_ASSERT( T == i_chip->getType() );
@@ -306,9 +307,19 @@ class VcmEvent : public TdEntry
      */
     uint32_t falseAlarm( STEP_CODE_DATA_STRUCT & io_sc );
 
+    /**
+     * @brief  Do extra analysis needed for Row Repair.
+     * @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 rowRepair( STEP_CODE_DATA_STRUCT & io_sc, bool & o_done );
+
   private: // instance variables
 
     const MemMark iv_mark; ///< The chip mark from hardware.
+
+    const bool iv_rowRepairEnabled; ///< True if Row Repair is enabled.
 };
 
 //------------------------------------------------------------------------------

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
@@ -170,6 +170,13 @@ uint32_t VcmEvent<TYPE_MBA>::startCmd()
 
     uint32_t stopCond = mss_MaintCmd::NO_STOP_CONDITIONS;
 
+    // Ensure we stop on each MCE if Row Repair is enabled.
+    if ( iv_rowRepairEnabled && (TD_PHASE_2 == iv_phase) )
+    {
+        stopCond |= mss_MaintCmd::STOP_ON_MCE;
+        stopCond |= mss_MaintCmd::STOP_IMMEDIATE;
+    }
+
     switch ( iv_phase )
     {
         case TD_PHASE_1:

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm_rt.C
@@ -211,6 +211,11 @@ uint32_t VcmEvent<TYPE_MBA>::startCmd()
 
     if ( TD_PHASE_2 == iv_phase ) stopCond |= mss_MaintCmd::STOP_ON_MCE;
 
+    // If Row Repair is enabled, we should stop immediately on error if there is
+    // an MCE during phase 2. This is already done because of the Centaur
+    // workarounds. This comment is just pointing out what should be done if for
+    // some reason the hardware is fixed.
+
     if ( iv_canResumeScrub )
     {
         // Resume the command from the next address to the end of this master

diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -57,6 +57,7 @@
 #include <fapi2_spd_access.H>
 #include <p9c_mss_maint_cmds.H>
 #include <prdfParserUtils.H>
+#include <p9c_mss_rowRepairFuncs.H>
 
 using namespace TARGETING;
 
@@ -565,6 +566,60 @@ uint32_t getMemAddrRange<TYPE_MCA>( ExtensibleChip * i_chip,
                                     MemAddr & o_startAddr, MemAddr & o_endAddr,
                                     uint8_t i_dimmSlct );
 
+//------------------------------------------------------------------------------
+
+template<>
+bool isRowRepairEnabled<TYPE_MBA>( ExtensibleChip * i_chip,
+                                   const MemRank & i_rank )
+{
+    #define PRDF_FUNC "[PlatServices::isRowRepairEnabled<TYPE_MBA>] "
+
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
+
+    bool o_isEnabled = false;
+
+    do
+    {
+        // Don't do row repair if DRAM repairs is disabled.
+        if ( areDramRepairsDisabled() ) break;
+
+        // The HWP will check both DIMMs on rank pair. So we could can use
+        // either one.
+        TargetHandleList list = getConnectedDimms( i_chip->getTrgt(), i_rank );
+        PRDF_ASSERT( !list.empty() );
+
+        TargetHandle_t dimm = list.front();
+
+        errlHndl_t errl = nullptr;
+        fapi2::Target<fapi2::TARGET_TYPE_DIMM> fapiDimm(dimm);
+        FAPI_INVOKE_HWP( errl, is_sPPR_supported, fapiDimm, o_isEnabled );
+        if ( nullptr != errl )
+        {
+            PRDF_ERR( PRDF_FUNC "is_sPPR_supported(0x%08x) failed",
+                      getHuid(dimm) );
+            PRDF_COMMIT_ERRL( errl, ERRL_ACTION_REPORT );
+            o_isEnabled = false; // just in case
+            break;
+        }
+
+    } while (0);
+
+    return o_isEnabled;
+
+    #undef PRDF_FUNC
+}
+
+template<>
+bool isRowRepairEnabled<TYPE_MCA>( ExtensibleChip * i_chip,
+                                   const MemRank & i_rank )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MCA == i_chip->getType() );
+
+    return false; // Not supported at this time.
+}
+
 //##############################################################################
 //##                    Nimbus Maintenance Command wrappers
 //##############################################################################

diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.H b/src/usr/diag/prdf/plat/prdfPlatServices.H
@@ -158,6 +158,15 @@ uint32_t getMemAddrRange( ExtensibleChip * i_chip,
                           VT & o_startAddr, VT & o_endAddr,
                           uint8_t i_dimmSlct = MAX_DIMM_PER_PORT );
 
+/**
+ * @brief  Queries if Row Repair is enabled.
+ * @param  i_chip An MCA or MBA.
+ * @param  i_rank Any rank on the target DIMMs.
+ * @return True if Row Repair is enabled on the DIMMs containing this rank.
+ *         False otherwise.
+ */
+template<TARGETING::TYPE T>
+bool isRowRepairEnabled( ExtensibleChip * i_chip, const MemRank & i_rank );
 
 //##############################################################################
 //##                  Nimbus/Centaur Maintenance Command wrappers

diff --git a/src/usr/diag/prdf/prdf_hb_only.mk b/src/usr/diag/prdf/prdf_hb_only.mk
@@ -182,6 +182,7 @@ prd_vpath += ${ROOTPATH}/src/import/chips/centaur/procedures/hwp/memory
 prd_obj_no_sim += p9c_mss_maint_cmds.o
 prd_obj_no_sim += p9c_dimmBadDqBitmapFuncs.o
 prd_obj_no_sim += p9c_query_channel_failure.o
+prd_obj_no_sim += p9c_mss_rowRepairFuncs.o
 
 ################################################################################
 # The following are hardware procedure utilities that we are pulling into the