PRD: rule file updates for MCA unit

Change-Id: Ib37b1ec290081428a9c627ec1683568382d3612f
RTC: 169104
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/36316
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@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/gerrit1/36515
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>

9 files changed, 199 insertions, 27 deletions

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
index 3c6b757fb..23e97f0d2 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.C
@@ -49,6 +49,51 @@ namespace MemEcc
 
 //------------------------------------------------------------------------------
 
+template<>
+void calloutMemUe<TYPE_MCA>( ExtensibleChip * i_chip, const MemRank & i_rank,
+                             STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[MemEcc::calloutMemUe] "
+
+    PRDF_ASSERT( TYPE_MCA == i_chip->getType() );
+
+    SCAN_COMM_REGISTER_CLASS * fir  = i_chip->getRegister( "DDRPHYFIR" );
+    int32_t l_rc = fir->Read();
+    if ( SUCCESS != l_rc )
+    {
+        PRDF_ERR( PRDF_FUNC "Read() failed on DDRPHYFIR: i_chip=0x%08x",
+                  i_chip->getHuid() );
+    }
+
+    // Check DDRPHYFIR[54:55,57:59] to determine if this UE is a side-effect.
+    if ( SUCCESS == l_rc && (0 != (fir->GetBitFieldJustified(54,6) & 0x37)) )
+    {
+        // Callout the MCA.
+        io_sc.service_data->SetCallout( i_chip->getTrgt() );
+    }
+    else
+    {
+        // Callout the rank anyway.
+        MemoryMru memmru ( i_chip->getTrgt(), i_rank,
+                           MemoryMruData::CALLOUT_RANK );
+        io_sc.service_data->SetCallout( memmru );
+    }
+
+    #undef PRDF_FUNC
+}
+
+template<>
+void calloutMemUe<TYPE_MBA>( ExtensibleChip * i_chip, const MemRank & i_rank,
+                             STEP_CODE_DATA_STRUCT & io_sc )
+{
+    PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
+
+    MemoryMru memmru ( i_chip->getTrgt(), i_rank, MemoryMruData::CALLOUT_RANK );
+    io_sc.service_data->SetCallout( memmru );
+}
+
+//------------------------------------------------------------------------------
+
 #ifdef __HOSTBOOT_RUNTIME
 
 template<TARGETING::TYPE T>
@@ -258,11 +303,9 @@ uint32_t analyzeFetchUe( ExtensibleChip * i_chip,
         D db = static_cast<D>(i_chip->getDataBundle());
         db->iv_ueTable.addEntry( UE_TABLE::FETCH_UE, addr );
 
-        // Callout the rank.
+        // Make the hardware callout.
         MemRank rank = addr.getRank();
-        MemoryMru memmru ( i_chip->getTrgt(), rank,
-                           MemoryMruData::CALLOUT_RANK );
-        io_sc.service_data->SetCallout( memmru );
+        calloutMemUe<T>( i_chip, rank, io_sc );
 
         #ifdef __HOSTBOOT_RUNTIME
 
diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H
index 58c1fa4f4..00dbd33e5 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemEccAnalysis.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -41,6 +41,17 @@ namespace MemEcc
 {
 
 /**
+ * @brief  Will check if the UE is a side-effect attention and make a callout
+ *         appropriately.
+ * @param  i_chip   MCA or MBA.
+ * @param  i_rank   Target rank.
+ * @param  io_sc    The step code data struct.
+ */
+template<TARGETING::TYPE T>
+void calloutMemUe( ExtensibleChip * i_chip, const MemRank & i_rank,
+                   STEP_CODE_DATA_STRUCT & io_sc );
+
+/**
  * @brief  Analyzes a fetch MPE attention.
  * @param  i_chip   MCA or MBA.
  * @param  i_rank   Target rank.
diff --git a/src/usr/diag/prdf/common/plat/mem/prdfP9Mca_common.C b/src/usr/diag/prdf/common/plat/mem/prdfP9Mca_common.C
index 5605bf32d..271856750 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfP9Mca_common.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfP9Mca_common.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -82,6 +82,87 @@ PRDF_PLUGIN_DEFINE( p9_mca, PostAnalysis );
 
 //##############################################################################
 //
+//                               DDRPHYFIR
+//
+//##############################################################################
+
+/**
+ * @brief  DDRPHYFIR[54:55,57:59] MCA/UE algorithm
+ * @param  i_chip MCA chip.
+ * @param  io_sc  The step code data struct.
+ * @return SUCCESS
+ */
+int32_t mcaUeAlgorithm( ExtensibleChip * i_chip,
+                        STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[p9_mca::mcaUeAlgorithm] "
+
+    SCAN_COMM_REGISTER_CLASS * fir = nullptr;
+    SCAN_COMM_REGISTER_CLASS * msk = nullptr;
+
+    // If the attention is currently at threshold or if there is a mainline or
+    // maintenance UE on at the same time as the attention:
+    //  - Make the error log predictive.
+    //  - Mask the attention.
+    //  - Do not clear the attention. This will be used during maintenance and
+    //    memory UE analysis to indicate that the MCA should be called out
+    //    instead of the DIMMs. This is unconventional process is needed because
+    //    maintenance UEs are always masked (handled manually in maintenance
+    //    command complete attentions) and memory UEs will get unmasked anytime
+    //    Targeted Diagnostics is complete on that area of memory. So we never
+    //    truly have a way to permanently mask the UEs.
+
+    bool maskDoNotClearAttn = io_sc.service_data->IsAtThreshold();
+
+    if ( !maskDoNotClearAttn )
+    {
+        fir = i_chip->getRegister("MCAECCFIR");
+        if ( SUCCESS != fir->Read() )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on MCAECCFIR: i_chip=0x%08x",
+                      i_chip->getHuid() );
+        }
+        else
+        {
+            maskDoNotClearAttn = fir->IsBitSet(14) || fir->IsBitSet(34);
+        }
+    }
+
+    if ( maskDoNotClearAttn )
+    {
+        // Get the active attentions of DDRPHYFIR[54:55,57:59] and mask.
+        fir = i_chip->getRegister("DDRPHYFIR");
+
+        if ( SUCCESS != fir->Read() )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on DDRPHYFIR: i_chip=0x%08x",
+                      i_chip->getHuid() );
+        }
+        else
+        {
+            uint64_t tmp = fir->GetBitFieldJustified(54, 6) & 0x37;
+
+            msk = i_chip->getRegister("DDRPHYFIR_MASK_OR");
+
+            msk->clearAllBits();
+            msk->SetBitFieldJustified( 54, 6, tmp );
+
+            if ( SUCCESS != msk->Write() )
+            {
+                PRDF_ERR( PRDF_FUNC "Write() failed on DDRPHYFIR_MASK_OR: "
+                          "i_chip=0x%08x", i_chip->getHuid() );
+            }
+        }
+    }
+
+    return maskDoNotClearAttn ? PRD_NO_CLEAR_FIR_BITS : SUCCESS;
+
+    #undef PRDF_FUNC
+}
+PRDF_PLUGIN_DEFINE( p9_mca, mcaUeAlgorithm );
+
+//##############################################################################
+//
 //                               MCAECCFIR
 //
 //##############################################################################
diff --git a/src/usr/diag/prdf/common/plat/p9/p9_common_actions.rule b/src/usr/diag/prdf/common/plat/p9/p9_common_actions.rule
index 45e65f0d5..564535bb5 100644
--- a/src/usr/diag/prdf/common/plat/p9/p9_common_actions.rule
+++ b/src/usr/diag/prdf/common/plat/p9/p9_common_actions.rule
@@ -5,7 +5,7 @@
 #
 # OpenPOWER HostBoot Project
 #
-# Contributors Listed Below - COPYRIGHT 2016
+# Contributors Listed Below - COPYRIGHT 2016,2017
 # [+] International Business Machines Corp.
 #
 #
@@ -83,6 +83,12 @@ actionclass threshold5phour
     threshold( field(5 / hour) );
 };
 
+/** Threshold of 5 per day */
+actionclass threshold5pday
+{
+    threshold( field(5 / day) );
+};
+
 ################################################################################
 #                       Threshold and Mask policy
 ################################################################################
diff --git a/src/usr/diag/prdf/common/plat/p9/p9_mca.rule b/src/usr/diag/prdf/common/plat/p9/p9_mca.rule
index 0fc5ec54b..787c8700a 100644
--- a/src/usr/diag/prdf/common/plat/p9/p9_mca.rule
+++ b/src/usr/diag/prdf/common/plat/p9/p9_mca.rule
@@ -414,7 +414,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14 )
     /** MCAECCFIR[17]
      *  Mainline read IUE
      */
-    (rMCAECCFIR, bit(17)) ? defaultMaskedError;
+    (rMCAECCFIR, bit(17)) ? mainline_iue_handling;
 
     /** MCAECCFIR[18]
      *  Mainline read IRCD
@@ -424,7 +424,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14 )
     /** MCAECCFIR[19]
      *  Mainline read IMPE
      */
-    (rMCAECCFIR, bit(19)) ? defaultMaskedError;
+    (rMCAECCFIR, bit(19)) ? mainline_impe_handling;
 
     /** MCAECCFIR[20:27]
      *  Maintenance MPE
@@ -479,7 +479,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14 )
     /** MCAECCFIR[37]
      *  Maintenance IUE
      */
-    (rMCAECCFIR, bit(37)) ? defaultMaskedError;
+    (rMCAECCFIR, bit(37)) ? maintenance_iue_handling;
 
     /** MCAECCFIR[38]
      *  Maintenance IRCD
@@ -489,7 +489,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14 )
     /** MCAECCFIR[39]
      *  Maintenance IMPE
      */
-    (rMCAECCFIR, bit(39)) ? defaultMaskedError;
+    (rMCAECCFIR, bit(39)) ? maintenance_impe_handling;
 
     /** MCAECCFIR[40]
      *  spare
@@ -625,12 +625,12 @@ group gDDRPHYFIR filter singlebit, cs_root_cause
     /** DDRPHYFIR[54]
      *  Non-recoverable FSM error
      */
-    (rDDRPHYFIR, bit(54)) ? defaultMaskedError;
+    (rDDRPHYFIR, bit(54)) ? mca_ue_algorithm_th_5perDay;
 
     /** DDRPHYFIR[55]
      *  Full bus impact Register Parity Error
      */
-    (rDDRPHYFIR, bit(55)) ? defaultMaskedError;
+    (rDDRPHYFIR, bit(55)) ? mca_ue_algorithm_th_1;
 
     /** DDRPHYFIR[56]
      *  DDRPHY Parity errors
@@ -640,17 +640,17 @@ group gDDRPHYFIR filter singlebit, cs_root_cause
     /** DDRPHYFIR[57]
      *  FSM errors
      */
-    (rDDRPHYFIR, bit(57)) ? defaultMaskedError;
+    (rDDRPHYFIR, bit(57)) ? mca_ue_algorithm_th_5perDay;
 
     /** DDRPHYFIR[58]
      *  Register parity error impacting 16 bits
      */
-    (rDDRPHYFIR, bit(58)) ? defaultMaskedError;
+    (rDDRPHYFIR, bit(58)) ? mca_ue_algorithm_th_1;
 
     /** DDRPHYFIR[59]
      *  Register parity error impacting 8 bits
      */
-    (rDDRPHYFIR, bit(59)) ? defaultMaskedError;
+    (rDDRPHYFIR, bit(59)) ? mca_ue_algorithm_th_1;
 
     /** DDRPHYFIR[60]
      *  Register PE 4 bit impact
diff --git a/src/usr/diag/prdf/common/plat/p9/p9_mca_actions.rule b/src/usr/diag/prdf/common/plat/p9/p9_mca_actions.rule
index 6579b1d84..f708196a8 100644
--- a/src/usr/diag/prdf/common/plat/p9/p9_mca_actions.rule
+++ b/src/usr/diag/prdf/common/plat/p9/p9_mca_actions.rule
@@ -5,7 +5,7 @@
 #
 # OpenPOWER HostBoot Project
 #
-# Contributors Listed Below - COPYRIGHT 2016
+# Contributors Listed Below - COPYRIGHT 2016,2017
 # [+] International Business Machines Corp.
 #
 #
@@ -76,3 +76,24 @@ actionclass rcd_parity_error
     funccall("RcdParityError");                # Run TPS on TH for all MCA ranks
 };
 
+actionclass mainline_iue_handling     { TBDDefaultCallout; }; # TODO RTC 165383
+actionclass mainline_impe_handling    { TBDDefaultCallout; }; # TODO RTC 165384
+actionclass maintenance_iue_handling  { TBDDefaultCallout; }; # TODO RTC 165383
+actionclass maintenance_impe_handling { TBDDefaultCallout; }; # TODO RTC 165384
+
+/** MCA/UE algroithm, threshold 5 per day */
+actionclass mca_ue_algorithm_th_5perDay
+{
+    calloutSelfMed;
+    threshold5pday;
+    funccall("mcaUeAlgorithm"); # must be called last
+};
+
+/** MCA/UE algroithm, threshold 1 */
+actionclass mca_ue_algorithm_th_1
+{
+    calloutSelfMed;
+    threshold1;
+    funccall("mcaUeAlgorithm"); # must be called last
+};
+
diff --git a/src/usr/diag/prdf/common/plat/p9/p9_mca_regs.rule b/src/usr/diag/prdf/common/plat/p9/p9_mca_regs.rule
index 0317b5914..cb16de1f0 100644
--- a/src/usr/diag/prdf/common/plat/p9/p9_mca_regs.rule
+++ b/src/usr/diag/prdf/common/plat/p9/p9_mca_regs.rule
@@ -30,6 +30,7 @@
     ############################################################################
     # P9 MCA target MCAECCFIR
     ############################################################################
+
     register MCAECCFIR_AND
     {
         name        "P9 MCA target MCAECCFIR AND";
@@ -39,6 +40,18 @@
     };
 
     ############################################################################
+    # P9 MCA target DDRPHYFIR
+    ############################################################################
+
+    register DDRPHYFIR_MASK_OR
+    {
+        name        "P9 MCA target DDRPHYFIR MASK OR";
+        scomaddr    0x07011005;
+        capture     group never;
+        access      write_only;
+    };
+
+    ############################################################################
     # P9 Hardware Mark Stores
     ############################################################################
     register HW_MS0
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
index 7490802fb..d17021c7a 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -30,6 +30,7 @@
 #include <prdfMemTdCtlr.H>
 
 // Platform includes
+#include <prdfMemEccAnalysis.H>
 #include <prdfMemMark.H>
 #include <prdfMemoryMru.H>
 #include <prdfMemScrubUtils.H>
@@ -137,8 +138,7 @@ uint32_t __checkEcc( ExtensibleChip * i_chip, const MemRank & i_rank,
             io_sc.service_data->setSignature(     huid, PRDFSIG_MaintUE );
 
             // Add the rank to the callout list.
-            MemoryMru mm { trgt, i_rank, MemoryMruData::CALLOUT_RANK };
-            io_sc.service_data->SetCallout( mm );
+            MemEcc::calloutMemUe<T>( i_chip, i_rank, io_sc );
 
             // Make the error log predictive.
             io_sc.service_data->setServiceCall();
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
index 00a230ca5..5ab4183bf 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemVcm_ipl.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -26,6 +26,7 @@
 /** @file prdfMemVcm_ipl.C */
 
 // Platform includes
+#include <prdfMemEccAnalysis.H>
 #include <prdfMemVcm.H>
 #include <prdfMemScrubUtils.H>
 #include <prdfPlatServices.H>
@@ -98,9 +99,7 @@ uint32_t VcmEvent<TYPE_MCA>::nextStep( STEP_CODE_DATA_STRUCT & io_sc,
                 }
 
                 //Add the rank to the callout list (via MemoryMru)
-                MemoryMru memmru(iv_chip->getTrgt(), iv_rank,
-                        MemoryMruData::CALLOUT_RANK);
-                io_sc.service_data->SetCallout( memmru );
+                MemEcc::calloutMemUe<TYPE_MCA>( iv_chip, iv_rank, io_sc );
 
                 //Make the error log predictive
                 io_sc.service_data->setServiceCall();
@@ -159,9 +158,7 @@ uint32_t VcmEvent<TYPE_MCA>::nextStep( STEP_CODE_DATA_STRUCT & io_sc,
                 }
 
                 //Add the rank to the callout list (via MemoryMru)
-                MemoryMru memmru(iv_chip->getTrgt(), iv_rank,
-                        MemoryMruData::CALLOUT_RANK);
-                io_sc.service_data->SetCallout( memmru );
+                MemEcc::calloutMemUe<TYPE_MCA>( iv_chip, iv_rank, io_sc );
 
                 //Make the error log predictive
                 io_sc.service_data->setServiceCall();