Add error handling to TOD code

Change-Id: I8485b7180aaccc6c8d7db1ba7279933b10d94435
RTC: 161602
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/41881
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Reviewed-by: Martin Gloff <mgloff@us.ibm.com>
Reviewed-by: Matthew A. Ploetz <maploetz@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>

6 files changed, 99 insertions, 134 deletions

diff --git a/src/usr/isteps/istep18/TodControls.C b/src/usr/isteps/istep18/TodControls.C
index c22f64984..e30844966 100755
--- a/src/usr/isteps/istep18/TodControls.C
+++ b/src/usr/isteps/istep18/TodControls.C
@@ -49,6 +49,7 @@
 #include <p9_perv_scom_addresses.H>
 #include <p9_tod_setup.H>
 #include <errl/errlentry.H>
+#include <errl/errludtarget.H>
 #include <p9_tod_utils.H>
 #include <isteps/tod_init_reasoncodes.H>
 
@@ -152,6 +153,7 @@ errlHndl_t TodControls ::pickMdmt(const p9_tod_setup_tod_sel i_config)
                                         GETHUID(l_procItr->getTarget()),
                                         l_drwItr->getId(),
                                         i_config);
+                                errlCommit(l_errHdl, TOD_COMP_ID);
                             }
                             break;
                         }
@@ -214,6 +216,7 @@ errlHndl_t TodControls ::pickMdmt(const p9_tod_setup_tod_sel i_config)
                             getAttr<TARGETING::ATTR_HUID>(),
                             l_pTodDrw->getId(),
                             i_config);
+                    errlCommit(l_errHdl, TOD_COMP_ID);
                 }
             }
         }
@@ -234,11 +237,37 @@ errlHndl_t TodControls ::pickMdmt(const p9_tod_setup_tod_sel i_config)
          * @custdesc     Host Processor Firmware couldn't detect any
          *               functional master processor required to boot the host
          */
+        const bool hbSwError = true;
         l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                           ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                            TOD_PICK_MDMT,
                            TOD_MASTER_TARGET_NOT_FOUND,
-                           i_config);
+                           i_config,
+                           hbSwError);
+
+        //Check the list of garded targets on the system and pick the garded
+        //targets for adding it into FFDC data.
+        //
+        std::vector<TARGETING::ATTR_HUID_type>::iterator l_iter;
+
+        for ( l_iter = iv_gardedTargets.begin();
+                l_iter != iv_gardedTargets.end();
+                ++l_iter )
+        {
+            //Get the target corresponding to the HUID stored in
+            //iv_gardedTargets
+            TARGETING::Target* l_pTarget =
+                 TARGETING::Target::getTargetFromHuid(*l_iter);
+            if ( l_pTarget )
+            {
+                if ( TARGETING::TYPE_PROC == GETTYPE(l_pTarget))
+                {
+                    // Add garded PROC targets into the errorlog
+                    ERRORLOG::ErrlUserDetailsTarget(l_pTarget,
+                        "GARDed Part").addToLog(l_errHdl);
+                }
+            }
+        }
     }
     else
     {
@@ -300,11 +329,13 @@ errlHndl_t  TodControls::buildTodDrawers(
               * @userdata1    system target's HUID
               * @devdesc      MDMT could not find a functional node
               */
+            const bool hbSwError = true;
             l_errHdl = new ERRORLOG::ErrlEntry(
-                               ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                               ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                                TOD_BUILD_TOD_DRAWERS,
                                TOD_NO_FUNC_NODE_AVAILABLE,
-                               l_pSysTarget->getAttr<ATTR_HUID>());
+                               l_pSysTarget->getAttr<ATTR_HUID>(),
+                               hbSwError);
             break;
         }
 
@@ -426,6 +457,11 @@ errlHndl_t  TodControls::buildTodDrawers(
                                TOD_BUILD_TOD_DRAWERS,
                                TOD_NO_DRAWERS,
                                i_config);
+
+             l_errHdl->addProcedureCallout(
+             HWAS::EPUB_PRC_FIND_DECONFIGURED_PART,
+             HWAS::SRCI_PRIORITY_LOW);
+
         }
 
     }while(0);
@@ -568,7 +604,7 @@ errlHndl_t TodControls ::queryActiveConfig(
              *               configuring Time Of Day on the Host processors
              */
             l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                           ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                            TOD_QUERY_ACTIVE_CONFIG,
                            TOD_NO_VALID_MDMT_FOUND);
             break;
@@ -683,11 +719,13 @@ errlHndl_t TodControls ::queryActiveConfig(
                  * @custdesc     Service Processor Firmware encountered an
                  *               internal error
                  */
+                const bool hbSwError = true;
                 l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                           ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                            TOD_QUERY_ACTIVE_CONFIG,
                            TOD_NO_MDMT_ON_ACTIVE_CONFIG,
-                           o_activeConfig);
+                           o_activeConfig,
+                           hbSwError);
                 break;
             }
         }//isTodRunning check
@@ -1460,13 +1498,9 @@ errlHndl_t TodControls::gardGetGardedUnits(
         {
             TARGETING::Target * l_pTarget = NULL;
 
-            l_err = getTargetFromPhysicalPath(l_iter.iv_targetId,
-                               l_pTarget);
-            if(l_err != NULL)
-            {
-                TOD_ERR("Error finding target");
-                break;
-            }
+            // getTargetFromPhysicalPath will either succeed or assert
+            getTargetFromPhysicalPath(l_iter.iv_targetId, l_pTarget);
+
             GardedUnit_t l_gardedUnit;
             memset(&l_gardedUnit,0,sizeof(GardedUnit_t));
             l_gardedUnit.iv_huid =
@@ -1526,39 +1560,22 @@ errlHndl_t TodControls::gardGetGardedUnits(
 // getTargetFromPhysicalPath
 //******************************************************************************
 
-errlHndl_t TodControls::getTargetFromPhysicalPath(
+void TodControls::getTargetFromPhysicalPath(
     const TARGETING::ATTR_PHYS_PATH_type &i_path,
           TARGETING::Target*&  o_pTarget)
 
 {
     TOD_ENTER("getTargetFromPhysicalPath");
-    errlHndl_t l_err = NULL;
     do
     {
         o_pTarget =
                 TARGETING::targetService().toTarget(i_path);
-        if(o_pTarget == NULL)
-        {
-            TOD_ERR("Error in getting target from entity path[%s]",
-                        getPhysicalPathString(i_path));
-            /*@
-            * @errortype
-            * @moduleid     TOD_CONTROLS_GET_TARGET
-            * @reasoncode   TOD_NULL_INPUT_TARGET
-            * @devdesc      Target ID is invalid
-            * @custdesc     Service Processor Firmware encountered an internal
-            *               error
-            */
-            l_err = new ERRORLOG::ErrlEntry(
-                         ERRORLOG::ERRL_SEV_INFORMATIONAL,
-                         TOD_CONTROLS_GET_TARGET,
-                         TOD_NULL_INPUT_TARGET);
-            break;
-        }
+        TOD_ERR_ASSERT(o_pTarget != NULL,
+                "Error in getting target from entity path[%s]",
+                getPhysicalPathString(i_path));
     }
     while(0);
     TOD_EXIT();
-    return l_err;
 }
 
 //******************************************************************************
@@ -1589,26 +1606,7 @@ errlHndl_t TodControls::getParent(const TARGETING::Target *i_pTarget,
 
     do
     {
-        // If i_pTarget is NULL then create an error log
-        if(!i_pTarget)
-        {
-            TOD_ERR("Input Target handle is null");
-
-            //Create error
-            /*@
-             * @errortype
-             * @moduleid     TOD_UTIL_MOD_GET_PARENT
-             * @reasoncode   TOD_NULL_INPUT_TARGET
-             * @devdesc      NULL Target is supplied as input
-             * @custdesc     Service Processor Firmware encountered an internal
-             *               error
-             */
-            l_errl = new ERRORLOG::ErrlEntry(
-                         ERRORLOG::ERRL_SEV_INFORMATIONAL,
-                         TOD_UTIL_MOD_GET_PARENT,
-                         TOD_NULL_INPUT_TARGET);
-            break;
-        }
+        TOD_ERR_ASSERT(i_pTarget != NULL, "Input Target handle is null");
 
         // If we have a valid target, check if it is system
         l_type = i_pTarget->getAttr<TARGETING::ATTR_TYPE>();
@@ -1635,11 +1633,13 @@ errlHndl_t TodControls::getParent(const TARGETING::Target *i_pTarget,
              *               error
              */
 
+            const bool hbSwError = true;
             l_errl = new ERRORLOG::ErrlEntry(
-                         ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                         ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                          TOD_UTIL_MOD_GET_PARENT,
                          TOD_INVALID_TARGET,
-                         GETHUID(i_pTarget));
+                         GETHUID(i_pTarget),
+                         hbSwError);
 
             break;
         }
@@ -1701,14 +1701,6 @@ errlHndl_t TodControls::getParent(const TARGETING::Target *i_pTarget,
                           GETHUID(i_pTarget),
                           TWO_UINT32_TO_UINT64(l_list.size(), i_class));
 
-         // Assert will R/R the FSP, set the R/R bit in this error
-         // to indicate that this error is responsible for the R/R
-         //l_errl->setRsrlBit(true);
-
-         TOD_ERR_ASSERT(0, "ERROR - Parent not found for the supplied target."
-                        "Supplied target's HUID: 0x%08X",
-                        GETHUID(i_pTarget));
-
     }
 
     TOD_EXIT();
diff --git a/src/usr/isteps/istep18/TodControls.H b/src/usr/isteps/istep18/TodControls.H
index 96ad14f6d..c17526667 100755
--- a/src/usr/isteps/istep18/TodControls.H
+++ b/src/usr/isteps/istep18/TodControls.H
@@ -828,13 +828,10 @@ private:
        * @param[out] o_Target
        * Pointer to Target
        *
-       * @return errlHndl_t
-       * Returns error handle if supplied physical path
-       * is not valid
-       * else NULL
+       * assert if supplied physical path is not valid
        */
 
-      errlHndl_t getTargetFromPhysicalPath(
+      void getTargetFromPhysicalPath(
           const TARGETING::ATTR_PHYS_PATH_type &i_path,
                 TARGETING::Target*&  o_pTarget);
 
diff --git a/src/usr/isteps/istep18/TodDrawer.C b/src/usr/isteps/istep18/TodDrawer.C
index c8c078e3e..4c6335bf4 100755
--- a/src/usr/isteps/istep18/TodDrawer.C
+++ b/src/usr/isteps/istep18/TodDrawer.C
@@ -199,11 +199,13 @@ errlHndl_t TodDrawer::findMasterProc(TodProc*& o_drawerMaster) const
              *               host
              */
 
+            const bool hbSwError = true;
             l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                           ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                            TOD_FIND_MASTER_PROC,
                            TOD_NO_MASTER_PROC,
-                           iv_todDrawerId);
+                           iv_todDrawerId,
+                           hbSwError);
         }
     }while(0);
 
@@ -237,7 +239,7 @@ void TodDrawer::getPotentialMdmts(
 {
     TOD_ENTER("TodDrawer::getPotentialMdmts");
     bool l_isGARDed = false;
-    errlHndl_t l_errHndl = NULL;
+    errlHndl_t l_errHdl = NULL;
 
     const TARGETING::Target* l_procTarget = NULL;
 
@@ -250,15 +252,17 @@ void TodDrawer::getPotentialMdmts(
         if ( !(TOD::isProcBlackListed(l_procTarget)) )
         {
             //Check if the target is not garded
-            l_errHndl = TOD::checkGardStatusOfTarget(l_procTarget,
+            l_errHdl = TOD::checkGardStatusOfTarget(l_procTarget,
                         l_isGARDed);
 
-            if(l_errHndl)
+            if(l_errHdl)
             {
                 TOD_ERR("Failed in checkGardStatusOfTarget() to get the "
                         " GARD state for the target 0x%.8x",
                         GETHUID(l_procTarget));
 
+                errlCommit(l_errHdl, TOD_COMP_ID);
+
                 //Ignore this target as the gard status for this target
                 //could not be obtained.
                 continue;
diff --git a/src/usr/isteps/istep18/TodHwpIntf.C b/src/usr/isteps/istep18/TodHwpIntf.C
index 9495c3ab4..6fede2578 100755
--- a/src/usr/isteps/istep18/TodHwpIntf.C
+++ b/src/usr/isteps/istep18/TodHwpIntf.C
@@ -85,23 +85,6 @@ errlHndl_t todSetupHwp(const p9_tod_setup_tod_sel i_topologyType)
                     l_pMDMT->getMasterType(),
                     l_pMDMT->getBusIn(), l_pMDMT->getBusOut());
 
-            /*@
-             * @errortype
-             * @moduleid     TOD_SETUP_HWP
-             * @reasoncode   TOD_HWP_EXECUTION_FAILED
-             * @userdata1    MDMT HUID
-             * @userdata2    Topology Type
-             * @devdesc      Failed to configure TOD toplogy, due to
-             *               error returned by p9_tod_setup.
-             * @custdesc     Service Processor Firmware encountered an
-             *               internal error
-             */
-            l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
-                           TOD_SETUP_HWP,
-                           TOD_HWP_EXECUTION_FAILED,
-                           GETHUID(l_pMDMT->getTarget()),
-                           i_topologyType);
             break;
         }
     }while(0);
@@ -135,11 +118,13 @@ errlHndl_t todSaveRegsHwp(const p9_tod_setup_tod_sel i_topologyType)
              * @custdesc     Host failed to boot because there was a problem
              *               configuring Time Of Day on the Host processors
              */
+            const bool hbSwError = true;
             l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                           ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                            TOD_SAVEREGS_HWP,
                            TOD_NO_VALID_MDMT_FOUND,
-                           i_topologyType);
+                           i_topologyType,
+                           hbSwError);
             break;
         }
 
@@ -159,21 +144,6 @@ errlHndl_t todSaveRegsHwp(const p9_tod_setup_tod_sel i_topologyType)
                     l_pMDMT->getMasterType(),
                     l_pMDMT->getBusIn(), l_pMDMT->getBusOut());
 
-            /*@
-             * @errortype
-             * @moduleid     TOD_SAVEREGS_HWP
-             * @reasoncode   TOD_HWP_EXECUTION_FAILED
-             * @userdata1    Topology Type
-             * @devdesc      Failed to save TOD configuration registers,
-             *               due to error returned by p9_tod_save_config.
-             * @custdesc     Service Processor Firmware encountered an
-             *               internal error
-             */
-            l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
-                           TOD_SAVEREGS_HWP,
-                           TOD_HWP_EXECUTION_FAILED,
-                           i_topologyType);
             break;
         }
     }while(0);
@@ -209,12 +179,14 @@ errlHndl_t todInitHwp()
              * @custdesc     Host failed to boot because there was a problem
              *               configuring Time Of Day on the Host processors
              */
+            const bool hbSwError = true;
             l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                           ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                            TOD_INIT_HWP,
                            TOD_NO_VALID_MDMT_FOUND,
                            EMOD_TOD_INIT_HWP,
-                           TOD_PRIMARY);
+                           TOD_PRIMARY,
+                           hbSwError);
             break;
         }
 
@@ -236,22 +208,6 @@ errlHndl_t todInitHwp()
                     GETHUID(l_pMDMT->getTarget()),
                     l_pMDMT->getMasterType());
 
-            /*@
-             * @errortype
-             * @moduleid     TOD_INIT_HWP
-             * @reasoncode   TOD_HWP_EXECUTION_FAILED
-             * @userdata1    MDMT HUID
-             * @devdesc      Failed to initialize TOD, due to
-             *               error returned by p9_tod_init.
-             * @custdesc     Service Processor Firmware encountered an
-             *               internal error
-             */
-            l_errHdl = new ERRORLOG::ErrlEntry(
-                           ERRORLOG::ERRL_SEV_INFORMATIONAL,
-                           TOD_INIT_HWP,
-                           TOD_HWP_EXECUTION_FAILED,
-                           GETHUID(l_pMDMT->getTarget()));
-
             l_failingTodProc = reinterpret_cast<TARGETING::Target*>(
                     l_fapiFailingProcTarget.get());
 
diff --git a/src/usr/isteps/istep18/TodSvc.C b/src/usr/isteps/istep18/TodSvc.C
index dcd8a5eb2..85e89a911 100755
--- a/src/usr/isteps/istep18/TodSvc.C
+++ b/src/usr/isteps/istep18/TodSvc.C
@@ -177,6 +177,7 @@ errlHndl_t TodSvc::todSetup()
                      "for secondary topology.");
             //Report the error as informational - loss of redundancy,
             //but no loss of TOD function.
+            errlCommit(l_errHdl, TOD_COMP_ID);
             break;
         }
 
@@ -188,7 +189,7 @@ errlHndl_t TodSvc::todSetup()
             TOD_ERR("TOD setup failure: failed to create secondary topology.");
             //Report the error as informational - loss of redundancy,
             //but no loss of TOD function.
-
+            errlCommit(l_errHdl, TOD_COMP_ID);
             break;
         }
         l_secondary.dumpTopology();
@@ -202,7 +203,7 @@ errlHndl_t TodSvc::todSetup()
             TOD_ERR("TOD setup failure: secondary topology setup HWP.");
             //Report the error as informational - loss of redundancy,
             //but no loss of TOD function.
-
+            errlCommit(l_errHdl, TOD_COMP_ID);
             break;
         }
 
@@ -377,6 +378,8 @@ errlHndl_t TodSvc::todInit()
         if( l_errHdl )
         {
             TOD_ERR("TOD initialization failed for primary topology : HWP");
+            l_errHdl->setSev(ERRORLOG::ERRL_SEV_UNRECOVERABLE);
+            errlCommit(l_errHdl, TOD_COMP_ID);
             break;
         }
 
diff --git a/src/usr/isteps/istep18/TodSvcUtil.C b/src/usr/isteps/istep18/TodSvcUtil.C
index 913120815..522962925 100755
--- a/src/usr/isteps/istep18/TodSvcUtil.C
+++ b/src/usr/isteps/istep18/TodSvcUtil.C
@@ -35,6 +35,7 @@
 #include <targeting/common/commontargeting.H>
 #include <targeting/common/util.H>
 #include <attributeenums.H>
+#include <errl/errlmanager.H>
 
 #include "TodSvcUtil.H"
 #include "TodUtils.H"
@@ -71,12 +72,23 @@ void calloutTodEndPoint( const TARGETING::Target* const i_pTodEndPoint,
      */
 
     io_errHdl = new ERRORLOG::ErrlEntry(
-                        ERRORLOG::ERRL_SEV_INFORMATIONAL,
+                        ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                         TOD_ENDPOINT_CALLOUT,
                         TOD_MASTER_PATH_ERROR,
                         EMOD_CALLOUT_TOD_ENDPOINT,
                         GETHUID(i_pTodEndPoint));
 
+    io_errHdl->addHwCallout(
+            i_pTodEndPoint,
+            HWAS::SRCI_PRIORITY_HIGH,
+            HWAS::DECONFIG,
+            HWAS::GARD_Fatal);
+
+    //Get the PEER TOD end point on the OSC and callout as low
+    io_errHdl->addProcedureCallout(
+            HWAS::EPUB_PRC_TOD_CLOCK_ERR,
+            HWAS::SRCI_PRIORITY_LOW);
+
     //Get the PEER TOD end point on the OSC and callout as low
     TARGETING::PredicateCTM  l_todEndPointPred(
             TARGETING::CLASS_UNIT,TARGETING::TYPE_TODCLK);
@@ -105,10 +117,10 @@ void calloutTodEndPoint( const TARGETING::Target* const i_pTodEndPoint,
 
     if ( !l_todEndPointList.empty())
     {
-         io_errHdl->addHwCallout((*l_todEndPointList.begin()),
-                 HWAS::SRCI_PRIORITY_LOW,
-                 HWAS::DECONFIG,
-                 HWAS::GARD_Fatal);
+        io_errHdl->addHwCallout((*l_todEndPointList.begin()),
+                HWAS::SRCI_PRIORITY_LOW,
+                HWAS::DECONFIG,
+                HWAS::GARD_Fatal);
 
     }
 
@@ -190,6 +202,7 @@ uint32_t getMaxProcsOnSystem()
     else
     {
         TOD_ERR("getMaxConfigParams() Failed");
+        errlCommit(l_errHdl, TOD_COMP_ID);
     }
 
     return (l_maxProcCount);