PRD: Updates to AUE/IAUE handling

Change-Id: I7d403cb29bbeb2d5f383a38816b579e71fc3dc0d
RTC: 173491
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/39851
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: Brian J. Stegmiller <bjs@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/39967
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>

6 files changed, 110 insertions, 16 deletions

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.C b/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.C
index e0bcd5bb6..29b9815d8 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.C
@@ -233,8 +233,6 @@ uint32_t getMemReadAddr<TYPE_MBA>( ExtensibleChip * i_chip,
 
 //------------------------------------------------------------------------------
 
-#ifdef __HOSTBOOT_MODULE
-
 template<>
 uint32_t getMemMaintAddr<TYPE_MCBIST>( ExtensibleChip * i_chip,
                                        MemAddr & o_addr )
@@ -286,6 +284,7 @@ uint32_t getMemMaintAddr<TYPE_MCA>( ExtensibleChip * i_chip, MemAddr & o_addr )
 }
 
 //------------------------------------------------------------------------------
+
 template<>
 uint32_t getMemMaintAddr<TYPE_MBA>( ExtensibleChip * i_chip, MemAddr & o_addr )
 {
@@ -319,6 +318,8 @@ uint32_t getMemMaintAddr<TYPE_MBA>( ExtensibleChip * i_chip, MemAddr & o_addr )
 
 //------------------------------------------------------------------------------
 
+#ifdef __HOSTBOOT_MODULE
+
 uint32_t getMcbistMaintPort( ExtensibleChip * i_mcbChip,
                              std::vector<ExtensibleChip *> & o_mcaList )
 {
diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.H b/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.H
index fcbe33af9..07808a990 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.H
+++ b/src/usr/diag/prdf/common/plat/mem/prdfMemAddress.H
@@ -172,8 +172,6 @@ template<TARGETING::TYPE T>
 uint32_t getMemReadAddr( ExtensibleChip * i_chip, MemAddr::ReadReg i_reg,
                          MemAddr & o_addr );
 
-#ifdef __HOSTBOOT_MODULE
-
 /**
  * @brief  Reads the maintenance address from hardware.
  *
@@ -194,6 +192,8 @@ uint32_t getMemReadAddr( ExtensibleChip * i_chip, MemAddr::ReadReg i_reg,
 template<TARGETING::TYPE T>
 uint32_t getMemMaintAddr( ExtensibleChip * i_chip, MemAddr & o_addr );
 
+#ifdef __HOSTBOOT_MODULE
+
 /**
  * @brief  Queries broadcast mode information and determines which of the MCBIST
  *         ports were targeted for the command.
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 cac225d23..22dd60554 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfP9Mca_common.C
+++ b/src/usr/diag/prdf/common/plat/mem/prdfP9Mca_common.C
@@ -272,6 +272,69 @@ PRDF_PLUGIN_DEFINE( p9_mca, AnalyzeImpe );
 
 //------------------------------------------------------------------------------
 
+/**
+ * @brief  MCAECCFIR[13,16] - Mainline AUE and IAUE
+ * @param  i_chip MCA chip.
+ * @param  io_sc  The step code data struct.
+ * @return SUCCESS
+ */
+int32_t AnalyzeFetchAueIaue( ExtensibleChip * i_chip,
+                             STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[p9_mca::AnalyzeFetchAueIaue] "
+
+    MemAddr addr;
+    if ( SUCCESS != getMemReadAddr<TYPE_MCA>(i_chip, MemAddr::READ_AUE_ADDR,
+                                             addr) )
+    {
+        PRDF_ERR( PRDF_FUNC "getMemReadAddr(0x%08x,READ_AUE_ADDR) failed",
+                  i_chip->getHuid() );
+    }
+    else
+    {
+        MemRank rank = addr.getRank();
+        MemoryMru mm { i_chip->getTrgt(), rank, MemoryMruData::CALLOUT_RANK };
+        io_sc.service_data->SetCallout( mm, MRU_HIGH );
+    }
+
+    return SUCCESS; // nothing to return to rule code
+
+    #undef PRDF_FUNC
+}
+PRDF_PLUGIN_DEFINE( p9_mca, AnalyzeFetchAueIaue );
+
+/**
+ * @brief  MCAECCFIR[33] - Maintenance AUE
+ * @param  i_chip MCA chip.
+ * @param  io_sc  The step code data struct.
+ * @return SUCCESS
+ */
+int32_t AnalyzeMaintAue( ExtensibleChip * i_chip,
+                         STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #define PRDF_FUNC "[p9_mca::AnalyzeMaintAue] "
+
+    MemAddr addr;
+    if ( SUCCESS != getMemMaintAddr<TYPE_MCA>(i_chip, addr) )
+    {
+        PRDF_ERR( PRDF_FUNC "getMemMaintAddr(0x%08x) failed",
+                  i_chip->getHuid() );
+    }
+    else
+    {
+        MemRank rank = addr.getRank();
+        MemoryMru mm { i_chip->getTrgt(), rank, MemoryMruData::CALLOUT_RANK };
+        io_sc.service_data->SetCallout( mm, MRU_HIGH );
+    }
+
+    return SUCCESS; // nothing to return to rule code
+
+    #undef PRDF_FUNC
+}
+PRDF_PLUGIN_DEFINE( p9_mca, AnalyzeMaintAue );
+
+//------------------------------------------------------------------------------
+
 } // end namespace p9_mca
 
 } // end namespace PRDF
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 cda13518f..2f62b13ac 100644
--- a/src/usr/diag/prdf/common/plat/p9/p9_mca.rule
+++ b/src/usr/diag/prdf/common/plat/p9/p9_mca.rule
@@ -407,7 +407,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14, 17, 37 )
     /** MCAECCFIR[13]
      *  Mainline read AUE
      */
-    (rMCAECCFIR, bit(13)) ? all_dimm_H_th_1;
+    (rMCAECCFIR, bit(13)) ? mainline_aue_iaue_handling;
 
     /** MCAECCFIR[14]
      *  Mainline read UE
@@ -422,7 +422,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14, 17, 37 )
     /** MCAECCFIR[16]
      *  Mainline read IAUE
      */
-    (rMCAECCFIR, bit(16)) ? all_dimm_H_th_1;
+    (rMCAECCFIR, bit(16)) ? mainline_aue_iaue_handling;
 
     /** MCAECCFIR[17]
      *  Mainline read IUE
@@ -472,7 +472,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14, 17, 37 )
     /** MCAECCFIR[33]
      *  Maintenance AUE
      */
-    (rMCAECCFIR, bit(33)) ? all_dimm_H_th_1;
+    (rMCAECCFIR, bit(33)) ? maintenance_aue_handling;
 
     /** MCAECCFIR[34]
      *  Maintenance UE
@@ -487,7 +487,7 @@ group gMCAECCFIR filter singlebit, cs_root_cause( 14, 17, 37 )
     /** MCAECCFIR[36]
      *  Maintenance IAUE
      */
-    (rMCAECCFIR, bit(36)) ? all_dimm_H_th_1;
+    (rMCAECCFIR, bit(36)) ? maintenance_iaue_handling;
 
     /** MCAECCFIR[37]
      *  Maintenance IUE
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 95d591c49..a18372c0f 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
@@ -23,14 +23,6 @@
 #
 # IBM_PROLOG_END_TAG
 
-/** Callout all connected DIMMs HIGH on first occurence. */
-actionclass all_dimm_H_th_1
-{
-    callout(connected(TYPE_DIMM,0), MRU_HIGH);
-    callout(connected(TYPE_DIMM,1), MRU_HIGH);
-    threshold1;
-};
-
 /** Verify Chip Mark */
 actionclass verify_chip_mark_0 { funccall("AnalyzeFetchMpe_0"); };
 actionclass verify_chip_mark_1 { funccall("AnalyzeFetchMpe_1"); };
@@ -95,6 +87,31 @@ actionclass maintenance_iue_handling
 
 actionclass impe_handling { funccall("AnalyzeImpe"); };
 
+/** Handle Mainline AUEs/IAUEs */
+actionclass mainline_aue_iaue_handling
+{
+    funccall("AnalyzeFetchAueIaue");
+    calloutSelfLow;
+    threshold1;
+};
+
+/** Handle Maintenance AUEs */
+actionclass maintenance_aue_handling
+{
+    funccall("AnalyzeMaintAue");
+    calloutSelfLow;
+    threshold1;
+};
+
+/** Handle Maintenance IAUEs */
+actionclass maintenance_iaue_handling
+{
+    callout(connected(TYPE_DIMM,0), MRU_HIGH);
+    callout(connected(TYPE_DIMM,1), MRU_HIGH);
+    calloutSelfLow;
+    threshold1;
+};
+
 /** MCA/UE algroithm, threshold 5 per day */
 actionclass mca_ue_algorithm_th_5perDay
 {
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
index 59600c116..6e0c3878b 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -314,6 +314,19 @@ uint32_t startBgScrub<TYPE_MCA>( ExtensibleChip * i_mcaChip,
     //       background scrubbing never stops.
     mss::mcbist::stop_conditions stopCond;
 
+    // AUEs are checkstop attentions. Unfortunately, MCBIST commands do not stop
+    // when the system checkstops. Therefore, we must set the stop condition for
+    // AUEs so that we can use the MCBMCAT register to determine where the error
+    // occurred. Note that there isn't a stop condition specifically for IAUEs.
+    // Instead, there is the RCE threshold. Unfortunately, the RCE counter is a
+    // combination of IUE, IAUE, IMPE, and IRCD errors. It is possible to use
+    // this threshold and simply restart background scrubbing each time there is
+    // an IUE, IMPE, or IRCD but there is concern that PRD might get stuck
+    // handling those attentions on every address even after thresholds have
+    // been reached. Therefore, we simplified the design and will simply call
+    // out both DIMMs for maintenance IAUEs.
+    stopCond.set_pause_on_aue(mss::ON);
+
     #ifdef CONFIG_HBRT_PRD
 
     stopCond.set_thresh_nce_int(1)