HTMGT: Save WOF reset reasons across all WOF resets

-Add query command to retrieve these reset reasons

Change-Id: I55436031b73eff95ec7e69020050dad556f4809a
RTC:192844
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/61903
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Martha Broyles <mbroyles@us.ibm.com>
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>

5 files changed, 105 insertions, 17 deletions

diff --git a/src/usr/htmgt/htmgt.C b/src/usr/htmgt/htmgt.C
index 83fcd4e51..4e6abbe77 100644
--- a/src/usr/htmgt/htmgt.C
+++ b/src/usr/htmgt/htmgt.C
@@ -727,6 +727,14 @@ namespace HTMGT
                         }
                         break;
 
+
+
+                    case PASSTHRU_WOF_RESET_REASONS:
+                        TMGT_INF("passThruCommand: Query WOF Reset Reasons");
+                        OccManager::getWOFResetReasons(o_rspLength, o_rspData);
+                        break;
+
+
                     case PASSTHRU_OCC_CFG_DATA:
                         if (i_cmdLength == 3)
                         {
diff --git a/src/usr/htmgt/htmgt_occ.C b/src/usr/htmgt/htmgt_occ.C
index 8c081f2f6..c5fd57885 100644
--- a/src/usr/htmgt/htmgt_occ.C
+++ b/src/usr/htmgt/htmgt_occ.C
@@ -60,6 +60,7 @@ namespace HTMGT
         iv_needsReset(false),
         iv_needsWofReset(false),
         iv_wofResetCount(0),
+        iv_wofResetReasons(0),
         iv_failed(false),
         iv_seqNumber(0),
         iv_homer(i_homer),
@@ -1336,7 +1337,7 @@ namespace HTMGT
             (0 == safeMode))
         {
             // Make sure OCCs were built first (so data is valid)
-            errlHndl_t err = _buildOccs(); // if not a already built.
+            errlHndl_t err = _buildOccs(); // if not already built.
             if (err)
             {
                 TMGT_ERR("_getOccData: failed to build OCC structures "
@@ -1396,6 +1397,35 @@ namespace HTMGT
         o_length = index;
     }
 
+    // Get WOF Reset reasons for all OCCs
+    // NOTE: Data returned is of the form [ 1-byte ID | 4-byte bit vector]
+    //       per OCC instance found
+    void OccManager::_getWOFResetReasons(uint16_t & o_length,
+                                         uint8_t * o_data)
+    {
+        uint16_t index = 0;
+
+        // Make sure OCCs were built first (so data is valid)
+        errlHndl_t l_err = _buildOccs(); // if not already built.
+        if( l_err )
+        {
+            TMGT_ERR("_getWOFResetReasons: Failed to build OCC structures "
+                     "rc=0x%04X", l_err->reasonCode());
+            ERRORLOG::errlCommit(l_err, HTMGT_COMP_ID);
+        }
+
+        // Iterate through OCC objects to get bit vectors.
+        for( const auto & occ : iv_occArray )
+        {
+            o_data[index++] = occ->getInstance();
+            UINT32_PUT(&o_data[index], occ->getWofResetReasons());
+            index += 4;
+        }
+
+        o_length = index;
+    }
+
+
 
     // Set default pstate table type and reset all OCCs to pick them up
     errlHndl_t OccManager::_loadPstates(bool i_normalPstates)
@@ -1456,36 +1486,39 @@ namespace HTMGT
         {
             sys->tryGetAttr<TARGETING::ATTR_HTMGT_SAFEMODE>(safeMode);
         }
-
         for( const auto & occ : iv_occArray )
         {
-            if (occ->iv_resetCount != 0)
+            if ( occ->iv_resetCount != 0 )
             {
                 TMGT_INF("_clearResetCounts: Clearing OCC%d reset count "
                          "(was %d)",
                          occ->getInstance(),
                          occ->iv_resetCount);
                 occ->iv_resetCount = 0;
-                if (safeMode)
-                {
-                    // Clear OCC flags (failed, commEstablished, etc)
-                    occ->postResetClear();
-                }
             }
 
-            if(occ->iv_wofResetCount != 0)
+            if( occ->iv_wofResetCount != 0 )
             {
                 occ->iv_wofResetCount = 0;
                 TMGT_INF("_clearResetCounts: Clearing OCC%d WOF reset count "
                          "( was %d)",
                          occ->getInstance(),
                          occ->iv_wofResetCount);
-                if(safeMode)
-                {
-                    // Clear OCC flags
-                    occ->postResetClear();
-                }
             }
+
+            if( occ->iv_wofResetReasons != 0 )
+            {
+                TMGT_INF("_clearResetCounts: Clearing OCC%d WOF reset reasons "
+                         "( was 0x%08x)",
+                         occ->getInstance(),
+                         occ->iv_wofResetReasons );
+            }
+            if( safeMode )
+            {
+                // Clear OCC flags (failed, commEstablished, etc)
+                occ->postResetClear();
+            }
+
         }
 
         if (iv_sysResetCount != 0)
@@ -1584,6 +1617,12 @@ namespace HTMGT
         Singleton<OccManager>::instance()._getOccData(o_length, o_data);
     }
 
+    void OccManager::getWOFResetReasons(uint16_t & o_length, uint8_t * o_data)
+    {
+        Singleton<OccManager>::instance()._getWOFResetReasons(o_length,
+                                                              o_data);
+    }
+
     errlHndl_t OccManager::loadPstates(bool i_normalPstates)
     {
         return Singleton<OccManager>::instance()._loadPstates(i_normalPstates);
diff --git a/src/usr/htmgt/htmgt_occ.H b/src/usr/htmgt/htmgt_occ.H
index 40f564fcf..c712fe526 100644
--- a/src/usr/htmgt/htmgt_occ.H
+++ b/src/usr/htmgt/htmgt_occ.H
@@ -276,6 +276,14 @@ namespace HTMGT
 
 
             /**
+             * @brief Returns a bit vector representing aggregate WOF
+             *        resets reasons
+             *
+             * @return WOF Reset reason bit vector
+             */
+            uint32_t getWofResetReasons() { return iv_wofResetReasons; }
+
+            /**
              * @brief Return OCCs present bits
              *
              * @return bitmask representing this OCC position
@@ -385,6 +393,8 @@ namespace HTMGT
             bool            iv_needsWofReset;
             // WOF reset count
             uint8_t         iv_wofResetCount;
+            // WOF reset reason. Aggregate across all WOF resets until cleared.
+            uint32_t        iv_wofResetReasons;
             // true if OCC failed
             bool            iv_failed;
             // Sequence number of last/current OCC command
@@ -604,6 +614,16 @@ namespace HTMGT
              */
             static void getOccData(uint16_t & o_length, uint8_t *o_data);
 
+            /**
+             * @brief Collect WOF reset reasons for ALL OCCs
+             *
+             * @param[out] o_length Length of data returned in o_data
+             * @param[out] o_data   Buffer containing all OCCs 32-bit
+             *                      WOF reset reason vector preceded by
+             *                      their instanceId
+             */
+            static void getWOFResetReasons(uint16_t & o_length,
+                                           uint8_t * o_data);
 
             /**
              * @brief Load specified pstate tables for all OCCs.
@@ -771,6 +791,9 @@ namespace HTMGT
             /** See getOccData() above */
             void _getOccData(uint16_t & o_length, uint8_t *o_data);
 
+            /** See getWOFResetReasons above */
+            void _getWOFResetReasons(uint16_t & o_length, uint8_t *o_data);
+
             /** See loadPstates() above */
             errlHndl_t _loadPstates(bool i_normalPstates);
 
diff --git a/src/usr/htmgt/htmgt_utility.H b/src/usr/htmgt/htmgt_utility.H
index e119a611a..6b16adfd7 100644
--- a/src/usr/htmgt/htmgt_utility.H
+++ b/src/usr/htmgt/htmgt_utility.H
@@ -174,6 +174,7 @@ namespace HTMGT
         //HOLD for future need.  PASSTHRU_QUERY_MODE_FUNCTION    = 0x07,
         PASSTHRU_ENA_DIS_OPAL_STATE     = 0x08,
         PASSTHRU_SET_OCC_STATE          = 0x09,
+        PASSTHRU_WOF_RESET_REASONS      = 0x0A,
         PASSTHRU_OCC_CFG_DATA           = 0x0C,
     };
 
diff --git a/src/usr/htmgt/occError.C b/src/usr/htmgt/occError.C
index 15e8a48bb..e491009d6 100644
--- a/src/usr/htmgt/occError.C
+++ b/src/usr/htmgt/occError.C
@@ -109,6 +109,11 @@ namespace HTMGT
 
             TMGT_BIN("OCC ELOG", l_occElog, 256);
 
+
+            // Get user details section
+            const occErrlUsrDtls_t *l_usrDtls_ptr = (occErrlUsrDtls_t *)
+                ((uint8_t*)l_occElog + sizeof(occErrlEntry_t));
+
             const uint32_t l_occSrc = OCCC_COMP_ID | l_occElog->reasonCode;
             ERRORLOG::errlSeverity_t severity =
                 ERRORLOG::ERRL_SEV_INFORMATIONAL;
@@ -130,13 +135,27 @@ namespace HTMGT
             bool l_occReset = false;
             elogProcessActions(l_occElog->actions, l_occReset, severity);
 
+
+
+            // Need to add WOF reason code to OCC object regardless of
+            // whether WOF resets are disabled.
+            if( l_occElog->actions & TMGT_ERRL_ACTIONS_WOF_RESET_REQUIRED )
+            {
+                iv_wofResetReasons |= l_usrDtls_ptr->userData1;
+                TMGT_ERR("WOF Reset Reasons for OCC%d = 0x%08x",
+                        iv_instance,
+                        iv_wofResetReasons);
+
+            }
+
             // Check if we need a WOF requested reset
             if(iv_needsWofReset == true)
             {
                 TMGT_ERR("WOF Reset detected! SRC = 0x%X",
                         l_occSrc);
+
                 // We compare against one less than the threshold because
-                // the WOF reset count doesnt get incremented until resetPrep
+                // the WOF reset count doesn't get incremented until resetPrep
                 if( iv_wofResetCount < (WOF_RESET_COUNT_THRESHOLD-1) )
                 {
                     // Not at WOF reset threshold yet. Set sev to INFO
@@ -156,8 +175,6 @@ namespace HTMGT
             //       srcs which have similar uniqueness
             // NOTE: SRC tags are NOT required here as these logs will get
             //       parsed with the OCC src tags
-            const occErrlUsrDtls_t *l_usrDtls_ptr = (occErrlUsrDtls_t *)
-                ((uint8_t*)l_occElog + sizeof(occErrlEntry_t));
             bldErrLog(l_errlHndl,
                       (htmgtModuleId)(l_usrDtls_ptr->modId & 0x00FF),
                       (htmgtReasonCode)l_occSrc, // occ reason code