PRD: additional FFDC requested for MCI timeouts

Also made updates to the MCIFIR based on version 59 of the RAS
spreadsheet.

Change-Id: I7587281d33147d12e2063910a2bad94e3c45e9c3
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/6540
Tested-by: Jenkins Server
Reviewed-by: Sachin Gupta <sgupta2m@in.ibm.com>
Reviewed-by: Christopher T. Phan <cphan@us.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>
Reviewed-by: Zane Shelley <zshelle@us.ibm.com>
Reviewed-on: http://gfw160.austin.ibm.com:8080/gerrit/6780

4 files changed, 46 insertions, 17 deletions

diff --git a/src/usr/diag/prdf/common/framework/rule/prdrCompile.C b/src/usr/diag/prdf/common/framework/rule/prdrCompile.C
index 81ebfd86c..450d6ab7f 100755
--- a/src/usr/diag/prdf/common/framework/rule/prdrCompile.C
+++ b/src/usr/diag/prdf/common/framework/rule/prdrCompile.C
@@ -496,11 +496,12 @@ uint32_t prdrActionArgMap(const std::string & i_arg)
         g_ActionArgMap[tmpStr] = errType;
 
 #ifdef __HOSTBOOT_MODULE
-        // FIXME: Hostboot currently does not support dump contents. This is a
-        // temporary fix.
-        g_ActionArgMap["DUMP_CONTENT_HW"]           = 0;
-        g_ActionArgMap["DUMP_CONTENT_SH"]           = 1;
-        g_ActionArgMap["DUMP_CONTENT_HWSUPERNOVA"]  = 2;
+        // FIXME: RTC 88391 Hostboot currently does not support dump contents.
+        //        This is a temporary fix.
+        g_ActionArgMap["DUMP_CONTENT_SW"]           = 0x80000000;
+        g_ActionArgMap["DUMP_CONTENT_HW"]           = 0x40000000;
+        g_ActionArgMap["DUMP_CONTENT_SH"]           = 0x20000000;
+        g_ActionArgMap["DUMP_CONTENT_HWSUPERNOVA"]  = 0x00001000;
 #else
         // Initialize Dump values. //@ecdf
         #include <hdctContent.H>
diff --git a/src/usr/diag/prdf/common/plat/pegasus/CommonActions.rule b/src/usr/diag/prdf/common/plat/pegasus/CommonActions.rule
index 252d2c6c4..6604b6aa9 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/CommonActions.rule
+++ b/src/usr/diag/prdf/common/plat/pegasus/CommonActions.rule
@@ -231,7 +231,6 @@ actionclass callout2ndLvlMedThr1
 {
     callout2ndLvlMed;
     threshold1;
-
 };
 
 ################################################################################
@@ -244,6 +243,12 @@ actionclass dumpSH
     dump(DUMP_CONTENT_SH);
 };
 
+/** Dump SW */
+actionclass dumpSW
+{
+    dump(DUMP_CONTENT_SW);
+};
+
 ################################################################################
 #                               Default callouts                               #
 ################################################################################
diff --git a/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule b/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule
index 579d2e2d2..372cd274c 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule
+++ b/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule
@@ -113,6 +113,16 @@ chip Mcs
         capture     req nonzero("MCIFIR");
     };
 
+    register MCIFIR_WOF
+    {
+        name        "MC0.MCS0.RIGHT.MCI.SCOMFIR.MCIFIRWOF";
+        scomaddr    0x02011848;
+        capture     type secondary;
+        capture     group default;
+        capture     group FirRegs;
+        capture     req nonzero("MCIFIR");
+    };
+
     ############################################################################
     # Error Report Registers
     ############################################################################
@@ -148,9 +158,9 @@ chip Mcs
 
 ################################################################################
 # PB Chiplet MCIFIR
+# Reviewed p8dd1_mss_FFDC_59.xls on 10/07/13
 ################################################################################
-# MCIFIR updated based on p8dd1_mss_FFDC_37_ reviewd.xls
-###############################################################################
+
 rule MciFir
 {
     CHECK_STOP:  MCIFIR & ~MCIFIR_MASK & ~MCIFIR_ACT0 & ~MCIFIR_ACT1;
@@ -197,7 +207,7 @@ group gMciFir attntype CHECK_STOP, RECOVERABLE, SPECIAL, UNIT_CS
     /** MCIFIR[6]
      *  MCIFIRQ_REPLAY_BUFFER_ECC_UE
      */
-    (MciFir, bit(6)) ? SelfHighThr1 ;
+    (MciFir, bit(6)) ? SelfHighThr1;
 
     /** MCIFIR[7]
      *  MCIFIRQ_MCI_CHINIT_STATE_MACHINE_TIMEOUT
@@ -309,7 +319,7 @@ group gMciFir attntype CHECK_STOP, RECOVERABLE, SPECIAL, UNIT_CS
     /** MCIFIR[26]
      *  MCIFIRQ_POWERBUS_PROTOCOL_ERROR
      */
-    (MciFir, bit(26)) ? callout2ndLvlMedThr1;
+    (MciFir, bit(26)) ? callout2ndLvlDumpSW;
 
     /** MCIFIR[27]
      *  MCIFIRQ_MCS_COMMAND_LIST_TIMEOUT_DUE_TO_POWERBUS
@@ -348,7 +358,7 @@ group gMciFir attntype CHECK_STOP, RECOVERABLE, SPECIAL, UNIT_CS
     /** MCIFIR[33]
      *  MCIFIRQ_NONFOREIGN_ACCESS_TO_FOREIGN_BAR
      */
-    (MciFir, bit(33)) ? callout2ndLvlMedThr1;
+    (MciFir, bit(33)) ? callout2ndLvlDumpSW;
 
     /** MCIFIR[34]
      *  MCIFIRQ_CENTAUR_SYNC_COMMAND_DETECTED
@@ -373,18 +383,18 @@ group gMciFir attntype CHECK_STOP, RECOVERABLE, SPECIAL, UNIT_CS
     /** MCIFIR[38]
      *  MCIFIRQ_HA_ILLEGAL_CONSUMER_ACCESS_ERROR
      */
-    (MciFir, bit(38)) ? callout2ndLvlMedThr1;
+    (MciFir, bit(38)) ? callout2ndLvlDumpSW;
 
     /** MCIFIR[39]
      *  MCIFIRQ_HA_ILLEGAL_PRODUCER_ACCESS_ERROR
      */
-    (MciFir, bit(39)) ? callout2ndLvlMedThr1;
+    (MciFir, bit(39)) ? callout2ndLvlDumpSW;
 
     /** MCIFIR[40]
      *  CHANNEL TIMEOUT ERROR
      */
-    # FIXME RTC 23127 No decision on MRU  as yet
-    (MciFir, bit(40)) ? defaultMaskedError;
+    # FIXME RTC 23127 No decision on MRU as yet
+    (MciFir, bit(40)) ? TBDDefaultCallout;
 
     /** MCIFIR[41]
      *  CENTAUR FAULT LINE
@@ -480,7 +490,17 @@ actionclass CalloutDmiBusAndThr2pd
     threshold2pday;
 };
 
+/** If there is a lane repair attention, do nothing. Otherwise, callout DMI bus
+    with threshold of 2/day. */
 actionclass analyzeSpareBitAndThr
 {
     try( funccall("checkSpareBit"), CalloutDmiBusAndThr2pd );
 };
+
+/** Callout 2nd Level and dump SW */
+actionclass callout2ndLvlDumpSW
+{
+    callout2ndLvlMedThr1;
+    dumpSW;
+};
+
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C b/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C
index 02af45f66..ca520f70c 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C
@@ -192,12 +192,15 @@ int32_t checkSpareBit( ExtensibleChip * i_mcsChip,
                        STEP_CODE_DATA_STRUCT & i_sc )
 {
     using namespace LaneRepair;
-    int32_t l_rc = FAIL;
+
+    int32_t l_rc = FAIL; // Default is to handle the attention (via rule code).
 
     ExtensibleChip * mbChip = getMcsDataBundle( i_mcsChip )->getMembChip();
 
-    if ( true == isSpareBitOnDMIBus( i_mcsChip, mbChip ))
+    if ( isSpareBitOnDMIBus(i_mcsChip, mbChip) )
     {
+        // Ignore attention and do not commit the error log.
+        i_sc.service_data->DontCommitErrorLog();
         l_rc = SUCCESS;
     }