HTMGT add attempt to reset OCC when OCC Activate fails

Change-Id: I964d2b68216c3ddabae73ce3b851bbc468ec96a7
RTC: 123180
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/16145
Reviewed-by: Christopher Cain <cjcain@us.ibm.com>
Tested-by: Jenkins Server
Reviewed-by: Matt Spinler <spinler@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

5 files changed, 118 insertions, 50 deletions

diff --git a/src/include/usr/htmgt/htmgt_reasoncodes.H b/src/include/usr/htmgt/htmgt_reasoncodes.H
index ade192d13..6fe269de8 100644
--- a/src/include/usr/htmgt/htmgt_reasoncodes.H
+++ b/src/include/usr/htmgt/htmgt_reasoncodes.H
@@ -48,6 +48,7 @@ namespace HTMGT
         HTMGT_MOD_CHECK_OCC_RSP         = 0x92,
         HTMGT_MOD_PARSE_OCC_RSP         = 0x94,
         HTMGT_MOD_HANLDE_OCC_EXCEPTION  = 0xE0,
+        HTMGT_MOD_ENABLE_OCC_ACTUATION  = 0xE1,
     };
 
     enum htmgtReasonCode
diff --git a/src/usr/htmgt/htmgt.C b/src/usr/htmgt/htmgt.C
index aff25008a..a2f556f94 100644
--- a/src/usr/htmgt/htmgt.C
+++ b/src/usr/htmgt/htmgt.C
@@ -184,42 +184,30 @@ namespace HTMGT
 
         if (NULL != l_err)
         {
-            TMGT_ERR("OCCs not all active.  System will stay in safe mode");
+            TMGT_ERR("OCCs not all active.  Attempting OCC Reset");
             TMGT_CONSOLE("OCCs are not active (rc=0x%04X). "
-                         "System will remain in safe mode",
+                         "Attempting OCC Reset",
                          l_err->reasonCode());
-            TMGT_INF("Calling HBOCC::stopAllOCCs");
-            errlHndl_t err2 = HBOCC::stopAllOCCs();
+            TMGT_INF("Calling resetOccs");
+            errlHndl_t err2 = OccManager::resetOccs(NULL);
             if(err2)
             {
-                TMGT_ERR("stopAllOCCs() failed with 0x%04X",
+                TMGT_ERR("OccManager:;resetOccs failed with 0x%04X",
                          err2->reasonCode());
-                ERRORLOG::errlCommit(err2, HTMGT_COMP_ID);
-            }
-
-            // Update error log to unrecoverable and set SRC
-            // to indicate the system will remain in safe mode
-            /*@
-             * @errortype
-             * @reasoncode      HTMGT_RC_OCC_CRIT_FAILURE
-             * @moduleid        HTMGT_MOD_LOAD_START_STATUS
-             * @userdata1[0:7]  load/start completed
-             * @devdesc         OCCs did not all reach active state,
-             *                  system will be in Safe Mode
-             */
-            bldErrLog(l_err, HTMGT_MOD_LOAD_START_STATUS,
-                      HTMGT_RC_OCC_CRIT_FAILURE,
-                      i_startCompleted, 0, 0, 1,
-                      ERRORLOG::ERRL_SEV_UNRECOVERABLE);
 
-            // Add level 2 support callout
-            l_err->addProcedureCallout(HWAS::EPUB_PRC_LVL_SUPP,
-                                       HWAS::SRCI_PRIORITY_MED);
-            // Add HB firmware callout
-            l_err->addProcedureCallout(HWAS::EPUB_PRC_HB_CODE,
-                                       HWAS::SRCI_PRIORITY_MED);
+                // Set original error log  as unrecoverable and commit
+                l_err->setSev(ERRORLOG::ERRL_SEV_UNRECOVERABLE);
+                ERRORLOG::errlCommit(l_err, HTMGT_COMP_ID);
 
-            ERRORLOG::errlCommit(l_err, HTMGT_COMP_ID);
+                // Commit occReset error
+                ERRORLOG::errlCommit(err2, HTMGT_COMP_ID);
+            }
+            else
+            {
+                // retry worked - commit original error as informational
+                l_err->setSev(ERRORLOG::ERRL_SEV_INFORMATIONAL);
+                ERRORLOG::errlCommit(l_err, HTMGT_COMP_ID);
+            }
         }
 
     } // end processOccStartStatus()
@@ -229,6 +217,19 @@ namespace HTMGT
     // Notify HTMGT that an OCC has an error to report
     void processOccError(TARGETING::Target * i_procTarget)
     {
+        TARGETING::Target* sys = NULL;
+        TARGETING::targetService().getTopLevelTarget(sys);
+        uint8_t safeMode = 0;
+
+        // If the system is in safemode then can't talk to OCCs -
+        // ignore call to processOccError
+        if(sys &&
+           sys->tryGetAttr<TARGETING::ATTR_HTMGT_SAFEMODE>(safeMode) &&
+           safeMode)
+        {
+            return;
+        }
+
         bool polledOneOcc = false;
         OccManager::buildOccs();
 
@@ -347,29 +348,57 @@ namespace HTMGT
     // Set the OCC state
     errlHndl_t enableOccActuation(bool i_occActivation)
     {
-        occStateId targetState = OCC_STATE_ACTIVE;
-        if (false == i_occActivation)
-        {
-            targetState = OCC_STATE_OBSERVATION;
-        }
+        errlHndl_t l_err = NULL;
+        TARGETING::Target* sys = NULL;
+
+        TARGETING::targetService().getTopLevelTarget(sys);
+        uint8_t safeMode = 0;
 
-        // Set state for all OCCs
-        errlHndl_t l_err = OccManager::setOccState(targetState);
-        if (NULL == l_err)
+        // If the system is in safemode then can't talk to OCCs -
+        // ignore call to enableOccActuation
+        if(sys &&
+           sys->tryGetAttr<TARGETING::ATTR_HTMGT_SAFEMODE>(safeMode) &&
+           safeMode)
         {
-            TMGT_INF("enableOccActuation: OCC states updated to 0x%02X",
-                     targetState);
+            /*@
+             * @errortype
+             * @reasoncode      HTMGT_RC_OCC_CRIT_FAILURE
+             * @moduleid        HTMGT_MOD_ENABLE_OCC_ACTUATION
+             * @userdata1[0:7]  OCC activate [1==true][0==false]
+             * @devdesc         Invalid operation when OCCs are in safemode
+             */
+            bldErrLog(l_err,
+                      HTMGT_MOD_ENABLE_OCC_ACTUATION,
+                      HTMGT_RC_OCC_CRIT_FAILURE,
+                      i_occActivation, 0, 0, 1,
+                      ERRORLOG::ERRL_SEV_UNRECOVERABLE);
         }
-
-        if (OccManager::occNeedsReset())
+        else
         {
-            TMGT_ERR("enableOccActuation(): OCCs need to be reset");
-            // Don't pass failed target as OCC should have already
-            // been marked as failed during the poll.
-            errlHndl_t err2 = OccManager::resetOccs(NULL);
-            if(err2)
+            occStateId targetState = OCC_STATE_ACTIVE;
+            if (false == i_occActivation)
             {
-                ERRORLOG::errlCommit(err2, HTMGT_COMP_ID);
+                targetState = OCC_STATE_OBSERVATION;
+            }
+
+            // Set state for all OCCs
+            l_err = OccManager::setOccState(targetState);
+            if (NULL == l_err)
+            {
+                TMGT_INF("enableOccActuation: OCC states updated to 0x%02X",
+                         targetState);
+            }
+
+            if (OccManager::occNeedsReset())
+            {
+                TMGT_ERR("enableOccActuation(): OCCs need to be reset");
+                // Don't pass failed target as OCC should have already
+                // been marked as failed during the poll.
+                errlHndl_t err2 = OccManager::resetOccs(NULL);
+                if(err2)
+                {
+                    ERRORLOG::errlCommit(err2, HTMGT_COMP_ID);
+                }
             }
         }
 
diff --git a/src/usr/htmgt/htmgt_activate.C b/src/usr/htmgt/htmgt_activate.C
index 7f54d6dbe..4cb46f033 100644
--- a/src/usr/htmgt/htmgt_activate.C
+++ b/src/usr/htmgt/htmgt_activate.C
@@ -39,6 +39,7 @@
 
 #include <ipmi/ipmisensor.H>
 #include <sys/time.h>
+#include <console/consoleif.H>
 
 using namespace TARGETING;
 
@@ -163,6 +164,14 @@ namespace HTMGT
             l_err = occ->ipmiSensor(i_activate);
             if( l_err )
             {
+                TMGT_ERR("setOccActiveSensors failed. (OCC%d state:%d)",
+                         occ->getInstance(),
+                         i_activate);
+
+                TMGT_CONSOLE("setOccActiveSensors failed. (OCC%d state:%d)",
+                         occ->getInstance(),
+                         i_activate);
+
                 ERRORLOG::errlCommit(l_err, HTMGT_COMP_ID);
             }
         }
diff --git a/src/usr/htmgt/htmgt_occ.C b/src/usr/htmgt/htmgt_occ.C
index 8a539f446..6bbc0ae57 100644
--- a/src/usr/htmgt/htmgt_occ.C
+++ b/src/usr/htmgt/htmgt_occ.C
@@ -248,7 +248,8 @@ namespace HTMGT
     OccManager::OccManager()
         :iv_occMaster(NULL),
         iv_state(OCC_STATE_UNKNOWN),
-        iv_targetState(OCC_STATE_ACTIVE)
+        iv_targetState(OCC_STATE_ACTIVE),
+        iv_resetCount(0)
     {
     }
 
@@ -605,6 +606,21 @@ namespace HTMGT
             }
         }
 
+        if(false == _occNeedsReset())
+        {
+            // No occ target needs reset - increment system reset count
+            ++iv_resetCount;
+
+            TMGT_INF("resetOCCs: Incrementing system OCC reset count to %d",
+                     iv_resetCount);
+
+            if(iv_resetCount > OCC_RESET_COUNT_THRESHOLD)
+            {
+                atThreshold = true;
+            }
+
+        }
+
         uint64_t retryCount = OCC_RESET_COUNT_THRESHOLD;
         while(retryCount)
         {
@@ -663,7 +679,7 @@ namespace HTMGT
              */
             bldErrLog(err,
                       HTMTG_MOD_OCC_RESET,
-                      HTMGT_RC_OCC_RESET_THREHOLD,
+                      HTMGT_RC_OCC_CRIT_FAILURE,
                       0, 0, 0, 0,
                       ERRORLOG::ERRL_SEV_UNRECOVERABLE);
         }
@@ -673,6 +689,13 @@ namespace HTMGT
         {
             err->setSev(ERRORLOG::ERRL_SEV_UNRECOVERABLE);
 
+            // Add level 2 support callout
+            err->addProcedureCallout(HWAS::EPUB_PRC_LVL_SUPP,
+                                     HWAS::SRCI_PRIORITY_MED);
+            // Add HB firmware callout
+            err->addProcedureCallout(HWAS::EPUB_PRC_HB_CODE,
+                                     HWAS::SRCI_PRIORITY_MED);
+
             TARGETING::Target* sys = NULL;
             TARGETING::targetService().getTopLevelTarget(sys);
             uint8_t safeMode = 1;
@@ -683,8 +706,13 @@ namespace HTMGT
                sys->setAttr<TARGETING::ATTR_HTMGT_SAFEMODE>(safeMode);
             }
 
-            TMGT_ERR("_resetOccs: Safe Mode RC: 0x%04X (OCC%d)",
+            TMGT_ERR("_resetOccs: Safe Mode (RC: 0x%04X OCC%d)",
                      cv_safeReturnCode, cv_safeOccInstance);
+
+            TMGT_CONSOLE("OCCs are not active. The system will remain in "
+                         "safe mode (RC: 0x%04x  for OCC%d)",
+                         cv_safeReturnCode,
+                         cv_safeOccInstance);
         }
 
         return err;
diff --git a/src/usr/htmgt/htmgt_occ.H b/src/usr/htmgt/htmgt_occ.H
index dec19b883..5ac545ab7 100644
--- a/src/usr/htmgt/htmgt_occ.H
+++ b/src/usr/htmgt/htmgt_occ.H
@@ -507,6 +507,7 @@ namespace HTMGT
             occList_t               iv_occArray;
             occStateId              iv_state;
             occStateId              iv_targetState;
+            uint8_t                 iv_resetCount;
 
             /**
              * @brief SRC that caused system to enter safe mode