PRD: MFG files

Change-Id: I432274c75a06500551643aca0a78df6a911e4c80
RTC: 22859
Reviewed-on: http://gfw160.austin.ibm.com:8080/gerrit/4312
Tested-by: Jenkins Server
Reviewed-by: Zane Shelley <zshelle@us.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

7 files changed, 170 insertions, 83 deletions

diff --git a/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.C b/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.C
index 3fd188b63..72cf4c288 100755
--- a/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.C
+++ b/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.C
@@ -85,9 +85,9 @@ fapi::TargetType getFapiType( TARGETING::TargetHandle_t i_target )
 //##                     System Level Utility Functions
 //##############################################################################
 
-void getECIDString( TargetHandle_t i_target, const char * o_ecidStr )
+void getECIDString( TargetHandle_t i_target, char * o_ecidStr )
 {
-    o_ecidStr ="";
+    sprintf(o_ecidStr, "%s", "N/A");
     PRDF_ERR( "[PlatServices::getECIDString] Function not implemented yet" );
 }
 
diff --git a/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.H b/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.H
index 532a76534..911283d9d 100755
--- a/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.H
+++ b/src/usr/diag/prdf/common/framework/service/prdfPlatServices_common.H
@@ -90,7 +90,7 @@ bool isMemoryPreservingIpl();
  * @return  Returns ECID string for given target
  */
 void getECIDString( TARGETING::TargetHandle_t i_target,
-                    const char * o_ecidStr );
+                    char * o_ecidStr );
 
 /**
  * @brief Get a PRD timer value based on the current time.
diff --git a/src/usr/diag/prdf/common/framework/service/prdf_ras_services.C b/src/usr/diag/prdf/common/framework/service/prdf_ras_services.C
index 379680b7f..b613cb5f4 100755
--- a/src/usr/diag/prdf/common/framework/service/prdf_ras_services.C
+++ b/src/usr/diag/prdf/common/framework/service/prdf_ras_services.C
@@ -1300,8 +1300,7 @@ will also be removed. Need to confirm if this code is required anymore.
         // to pass errLog to PRDF_HWUDUMP
         // for FSP specific SRC handling in the future
 #ifndef __HOSTBOOT_MODULE
-        /* FIXME: not sure if we still need this in fips810?
-        MnfgTrace(esig); */
+        MnfgTrace(esig);
 #endif
 
         PRDF_GET_PLID(errLog, dumpPlid);
@@ -1340,8 +1339,7 @@ will also be removed. Need to confirm if this code is required anymore.
     {
 
 #ifndef __HOSTBOOT_MODULE
-        /* FIXME: not sure if we still need this in fips810?
-        MnfgTrace(esig); */
+        MnfgTrace(esig);
 #endif
 
         PRDF_DTRAC( PRDF_FUNC"generating a terminating, or MP Fatal SRC" );
@@ -1461,44 +1459,10 @@ will also be removed. Need to confirm if this code is required anymore.
     #undef PRDF_FUNC
 }
 
-void prdfGetTargetString(TargetHandle_t  i_pTargetHandle,
-                         char * o_chipName, uint32_t i_sizeOfChipName )
-{
-    //FIXME  waiting on alternate implementation of toString function in targeting
-    //FIXME Commenting out current usage of getEntityPathString
-    //char * l_entityPathString =NULL;
-    //uint32_t l_tempSize  =0;
-    do
-    {
-        if(NULL==i_pTargetHandle )
-        {
-#ifdef  __HOSTBOOT_MODULE
-            sprintf( o_chipName, "????, " );
-#else
-            snprintf( o_chipName, i_sizeOfChipName, "????, " );
-#endif
-
-        }
-        else
-        {
-/*
-            l_entityPathString =PlatServices::getEntityPathString(i_pTargetHandle);
-            l_tempSize  = strlen(l_entityPathString );
-            if(l_tempSize < i_sizeOfChipName)
-                i_sizeOfChipName = l_tempSize;
-            memcpy(o_chipName ,l_entityPathString ,i_sizeOfChipName);
-            free(l_entityPathString);
-*/
-        }
-
-    } while (0);
-}
-
-
 // ----------------------------------------------------------------------------
 
 #ifndef __HOSTBOOT_MODULE
-void RasServices::MnfgTrace(ErrorSignature * l_esig )
+void ErrDataService::MnfgTrace(ErrorSignature * l_esig )
 {
     char  * MnfgFilename = NULL;
     uint32_t l_size = 0;
@@ -1506,43 +1470,49 @@ void RasServices::MnfgTrace(ErrorSignature * l_esig )
 
     if ( PlatServices::mfgMode() )
     {
-        errlHndl_t errorLog = UtilReg::path(MnfgKey,1,"prdfMfgErrors",MnfgFilename,l_size);
+        errlHndl_t errorLog = UtilReg::path(MnfgKey, 1, "prdfMfgErrors",
+                                            MnfgFilename, l_size);
         if (errorLog == NULL)
         {
             UtilFile l_mfgFile;
             l_mfgFile.Open(MnfgFilename,"a+");
 
-            char l_array[62];
-            char l_array2[42];
+            char l_string[100];
             uint32_t signature = l_esig->getSigId();
             HUID sigChip = l_esig->getChipId();
 
             // Get Entity Path String
             TargetHandle_t l_ptempHandle = PlatServices::getTarget(sigChip);
-            prdfGetTargetString(l_ptempHandle , l_array, 62);
-            l_mfgFile.write(l_array, strlen(l_array));
+            TARGETING::EntityPath path;
+            PlatServices::getEntityPath(l_ptempHandle, path,
+                                        EntityPath::PATH_PHYSICAL);
+            char *epStr = path.toString();
+            if (epStr)
+            {
+                snprintf(l_string, 100, "%s, ", path.toString());
+                free(epStr);
+            }
+
+            l_mfgFile.write(l_string, strlen(l_string));
 
             // Write Signature
-            snprintf(l_array, 62, "0x%08x,", signature);
-            l_mfgFile.write(l_array, 24);
+            snprintf(l_string, 100, "0x%08x, 0x%08x, ", sigChip, signature);
+            l_mfgFile.write(l_string, strlen(l_string));
 
             // Write chip ECID data
-            char ecidString[1024];
-            l_ptempHandle = PlatServices::getTarget(pfaData.PfaCalloutList[0].Callout);
-            //TODO TargetHandle conversion - not sure we need it now
-            PlatServices::getECIDString(l_ptempHandle , ecidString);
-            l_mfgFile.write(ecidString, strlen(ecidString));
+            PlatServices::getECIDString(l_ptempHandle, l_string);
+            l_mfgFile.write(l_string, strlen(l_string));
 
             // Write MRU list
             uint32_t n = 0;
             while ( (n < MruListLIMIT )  && (n < pfaData.PfaCalloutCount) )
             {
-                snprintf(l_array2, 16, ", %08x", pfaData.PfaCalloutList[n].Callout);
-                l_mfgFile.write(l_array2, 9);
+                snprintf(l_string, 100, " , %08x", pfaData.PfaCalloutList[n].Callout);
+                l_mfgFile.write(l_string, strlen(l_string));
                 ++n;
             }
-            snprintf(l_array2, 42, "\n");
-            l_mfgFile.write(l_array2, 1);
+            snprintf(l_string, 100, "\n");
+            l_mfgFile.write(l_string, 1);
 
             l_mfgFile.Close();
         }
diff --git a/src/usr/diag/prdf/common/framework/service/prdf_ras_services.H b/src/usr/diag/prdf/common/framework/service/prdf_ras_services.H
index 8a23966b8..c79a6213b 100755
--- a/src/usr/diag/prdf/common/framework/service/prdf_ras_services.H
+++ b/src/usr/diag/prdf/common/framework/service/prdf_ras_services.H
@@ -175,6 +175,16 @@ class ErrDataService
       */
     bool SdcRetrieve(sdcSaveFlagsEnum i_saveFlag, void * o_buffer);
 
+    /**
+     @brief Add Trace Data to File for Manufacturing.
+     @param[in] ErrorSignature
+     @return None.
+     @pre None.
+     @post None.
+     @exception None.
+     */
+    void MnfgTrace(ErrorSignature * l_esig);
+
     #endif // end if not __HOSTBOOT_MODULE
 
   private:
@@ -263,18 +273,6 @@ public:
   virtual errlHndl_t GenerateSrcPfa( ATTENTION_TYPE attn_type,
                                     ServiceDataCollector & sdc );
 
-#ifndef __HOSTBOOT_MODULE
-  /**
-   @brief Add Trace Data to File for Manufacturing.
-   @param[in] ErrorSignature
-   @return None.
-   @pre None.
-   @post None.
-   @exception None.
-   */
-  void MnfgTrace(ErrorSignature * l_esig);
-#endif
-
   static void SetTerminateOnCheckstop(bool setting)
   {
       ErrDataService::SetTerminateOnCheckstop(setting);
diff --git a/src/usr/diag/prdf/common/mnfgtools/prdfMfgThresholds.lst b/src/usr/diag/prdf/common/mnfgtools/prdfMfgThresholds.lst
index 30a8d89b0..d3d5ed944 100755
--- a/src/usr/diag/prdf/common/mnfgtools/prdfMfgThresholds.lst
+++ b/src/usr/diag/prdf/common/mnfgtools/prdfMfgThresholds.lst
@@ -1,3 +1,9 @@
+# P8 EX Thresholds
+P8EX_L2_CACHE_CES 1
+P8EX_L2_DIR_CES   1
+P8EX_L3_CACHE_CES 255
+P8EX_L3_DIR_CES   1
+
 # P8 Chip Thresholds
 P8CHIP_ONNODE_BUS_CES   1
 P8CHIP_OFFNODE_BUS_CES  1
diff --git a/src/usr/diag/prdf/common/plat/pegasus/Ex.rule b/src/usr/diag/prdf/common/plat/pegasus/Ex.rule
index 28f078d98..65596c9c3 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/Ex.rule
+++ b/src/usr/diag/prdf/common/plat/pegasus/Ex.rule
@@ -398,7 +398,7 @@ group gExChipletFir attntype CHECK_STOP, RECOVERABLE filter priority(3,5,6,7,4,2
     /** EX_CHIPLET_FIR[3]
      *  Attention from LFIR
      */
-    (ExChipletFir, bit(3))? analyze(gExLFir);
+    (ExChipletFir, bit(3))? analyzeExLFir;
 
     /** EX_CHIPLET_FIR[4]
      *  Attention from COREFIR
@@ -408,17 +408,17 @@ group gExChipletFir attntype CHECK_STOP, RECOVERABLE filter priority(3,5,6,7,4,2
     /** EX_CHIPLET_FIR[5]
      *  Attention from L2FIR
      */
-    (ExChipletFir, bit(5)) ? analyze(gL2Fir);
+    (ExChipletFir, bit(5)) ? analyzeL2Fir;
 
     /** EX_CHIPLET_FIR[6]
      *  Attention from L3FIR
      */
-    (ExChipletFir, bit(6)) ? analyze(gL3Fir);
+    (ExChipletFir, bit(6)) ? analyzeL3Fir;
 
     /** EX_CHIPLET_FIR[7]
      *  Attention from NCUFIR
      */
-    (ExChipletFir, bit(7)) ? analyze(gNcuFir);
+    (ExChipletFir, bit(7)) ? analyzeNcuFir;
 };
 
 ################################################################################
@@ -1347,6 +1347,7 @@ actionclass analyzeCore
     funccall("CheckCoreCheckstop");
     analyze(gCoreFir);
     funccall("MaskIfCoreCheckstop");
+    funccall("RestartTraceArray");
 };
 
 # Chiplet recoverable FIR indicated Core checkstop, so take actions for
@@ -1359,44 +1360,67 @@ actionclass analyzeCoreUnitCheckstop
     funccall("MaskIfCoreCheckstop");
 };
 
+actionclass analyzeExLFir
+{
+    analyze(gExLFir);
+    funccall("RestartTraceArray");
+};
+
+actionclass analyzeL2Fir
+{
+    analyze(gL2Fir);
+    funccall("RestartTraceArray");
+};
+
+actionclass analyzeL3Fir
+{
+    analyze(gL3Fir);
+    funccall("RestartTraceArray");
+};
+
+actionclass analyzeNcuFir
+{
+    analyze(gNcuFir);
+    funccall("RestartTraceArray");
+};
+
 actionclass L3DirCE
 {
     calloutSelfMed;
-    threshold32pday; #FIXME RTC:65507 add MFG threshold
+    threshold( field(32 / day), mfg_file(P8EX_L3_DIR_CES) );
 };
 
 actionclass L3UE
 {
     SelfMedThr1UE;
-    #funccall("L3UE"); #FIXME RTC:22859 support for L3 repairs and trace arrays
+    funccall("L3UE");
 };
 
 actionclass L3CE
 {
     calloutSelfMed;
-    threshold32pday; #FIXME RTC:65507 add MFG threshold
-    #funccall("L3CE"); #FIXME RTC:22859 support for L3 repairs and trace arrays
+    threshold( field(32 / day), mfg_file(P8EX_L3_CACHE_CES) );
+    funccall("L3CE");
 };
 
 
 actionclass L2CE
 {
     calloutSelfMed;
-    threshold32pday; #FIXME RTC:65507 add MFG threshold
-    #funcall("L2CE");#FIXME RTC 22859 Support for L2 Line Delete
-
+    threshold( field(32 / day), mfg_file(P8EX_L2_CACHE_CES) );
+    funccall("L2CE");
 };
 
 actionclass L2UE
 {
     SelfMedThr1UE;
-    #funcall("L2UE");#FIXME RTC 22859 Support for L2 Line Delete
+    funccall("L2UE");
 };
 
 actionclass L2DirCE
 {
     calloutSelfMed;
-    threshold32pday; #FIXME RTC:65507 add MFG threshold
+    threshold( field(32 / day), mfg_file(P8EX_L2_DIR_CES) );
 };
 
 actionclass L2ChipLevel2
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfP8Ex.C b/src/usr/diag/prdf/common/plat/pegasus/prdfP8Ex.C
index 196dcfda2..e453d375f 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfP8Ex.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfP8Ex.C
@@ -197,5 +197,94 @@ int32_t MaskIfCoreCheckstop( ExtensibleChip * i_chip,
     return l_rc;
 } PRDF_PLUGIN_DEFINE(Ex, MaskIfCoreCheckstop);
 
+/**
+ * @brief Restart Trace Arrays that have been stopped on error
+ * @param i_chip Ex chip.
+ * @param i_stepcode Step Code data struct
+ * @return PRD return code
+ */
+int32_t RestartTraceArray( ExtensibleChip * i_chip,
+                           STEP_CODE_DATA_STRUCT & i_stepcode )
+{
+    int32_t l_rc = SUCCESS;
+#ifndef __HOSTBOOT_MODULE
+    l_rc = PlatServices::RestartTraceArray(i_chip->GetChipHandle());
+#endif
+    return l_rc;
+} PRDF_PLUGIN_DEFINE(Ex, RestartTraceArray);
+
+/**
+ * @brief Handle an L3 UE
+ * @param i_chip Ex chip.
+ * @param i_stepcode Step Code data struct
+ * @return PRD return code
+ */
+int32_t L3UE( ExtensibleChip * i_chip,
+                           STEP_CODE_DATA_STRUCT & i_stepcode )
+{
+    //FIXME RTC:22859 support for L2/L3 repairs
+    int32_t l_rc = SUCCESS;
+#ifndef __HOSTBOOT_MODULE
+    l_rc = PlatServices::processL2L3TraceArray(i_chip->GetChipHandle(),
+                                               false);
+#endif
+    return l_rc;
+
+} PRDF_PLUGIN_DEFINE(Ex, L3UE);
+
+/**
+ * @brief Handle an L3 CE
+ * @param i_chip Ex chip.
+ * @param i_stepcode Step Code data struct
+ * @return PRD return code
+ */
+int32_t L3CE( ExtensibleChip * i_chip,
+                           STEP_CODE_DATA_STRUCT & i_stepcode )
+{
+    //FIXME RTC:22859 support for L2/L3 repairs
+    int32_t l_rc = SUCCESS;
+#ifndef __HOSTBOOT_MODULE
+    l_rc = PlatServices::processL2L3TraceArray(i_chip->GetChipHandle(),
+                                               false);
+#endif
+    return l_rc;
+} PRDF_PLUGIN_DEFINE(Ex, L3CE);
+
+/**
+ * @brief Handle an L2 UE
+ * @param i_chip Ex chip.
+ * @param i_stepcode Step Code data struct
+ * @return PRD return code
+ */
+int32_t L2UE( ExtensibleChip * i_chip,
+                           STEP_CODE_DATA_STRUCT & i_stepcode )
+{
+    //FIXME RTC:22859 support for L2/L3 repairs
+    int32_t l_rc = SUCCESS;
+#ifndef __HOSTBOOT_MODULE
+    l_rc = PlatServices::processL2L3TraceArray(i_chip->GetChipHandle(),
+                                               true);
+#endif
+    return l_rc;
+} PRDF_PLUGIN_DEFINE(Ex, L2UE);
+
+/**
+ * @brief Handle an L2 CE
+ * @param i_chip Ex chip.
+ * @param i_stepcode Step Code data struct
+ * @return PRD return code
+ */
+int32_t L2CE( ExtensibleChip * i_chip,
+                           STEP_CODE_DATA_STRUCT & i_stepcode )
+{
+    //FIXME RTC:22859 support for L2/L3 repairs
+    int32_t l_rc = SUCCESS;
+#ifndef __HOSTBOOT_MODULE
+    l_rc = PlatServices::processL2L3TraceArray(i_chip->GetChipHandle(),
+                                               true);
+#endif
+    return l_rc;
+} PRDF_PLUGIN_DEFINE(Ex, L2CE);
+
 } // end namespace Ex
 } // end namespace PRDF