HTMGT: Ensure processor callout added when going to safe mode

Change-Id: Ib5e1035121ed574a97416d59498b39f997131fd6
CQ: SW398232
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/50597
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Reviewed-by: Martha Broyles <mbroyles@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: Sheldon R. Bailey <baileysh@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>

3 files changed, 80 insertions, 16 deletions

diff --git a/src/usr/htmgt/htmgt_cfgdata.C b/src/usr/htmgt/htmgt_cfgdata.C
index 995042f4f..457aee9dd 100644
--- a/src/usr/htmgt/htmgt_cfgdata.C
+++ b/src/usr/htmgt/htmgt_cfgdata.C
@@ -212,7 +212,7 @@ namespace HTMGT
                                 {
                                     TMGT_ERR("sendOccConfigData: OCC%d cfg "
                                              "format 0x%02X had bad rsp status"
-                                             " 0x%02X for sysConfig",
+                                             " 0x%02X",
                                              occInstance, format,
                                              cmd.getRspStatus());
                                 }
@@ -761,7 +761,15 @@ void getSystemConfigMessageData(const TargetHandle_t i_occ, uint8_t* o_data,
     uint64_t index = 0;
     uint32_t SensorID1 = 0;
     uint32_t SensorID2 = 0;
-        assert(o_data != nullptr);
+    assert(o_data != nullptr);
+
+    TargetHandle_t sys = nullptr;
+    TargetHandleList nodes;
+    targetService().getTopLevelTarget(sys);
+    assert(sys != nullptr);
+    getChildAffinityTargets(nodes, sys, CLASS_ENC, TYPE_NODE);
+    assert(!nodes.empty());
+    TargetHandle_t node = nodes[0];
 
     o_data[index++] = OCC_CFGDATA_SYS_CONFIG;
     o_data[index++] = OCC_CFGDATA_SYS_CONFIG_VERSION;
@@ -781,6 +789,10 @@ void getSystemConfigMessageData(const TargetHandle_t i_occ, uint8_t* o_data,
     TargetHandleList cores;
     getChildChiplets(cores, proc, TYPE_CORE, false);
 
+    TMGT_INF("getSystemConfigMessageData: systemType: 0x%02X, "
+             "procSensor: 0x%04X, %d cores, %d nodes",
+             G_opalMode, SensorID1, cores.size(), nodes.size());
+
     for (uint64_t core=0; core<CFGDATA_CORES; core++)
     {
         SensorID1 = 0;
@@ -804,16 +816,6 @@ void getSystemConfigMessageData(const TargetHandle_t i_occ, uint8_t* o_data,
         index += 4;
     }
 
-
-    TargetHandle_t sys = nullptr;
-    TargetHandleList nodes;
-    targetService().getTopLevelTarget(sys);
-    assert(sys != nullptr);
-    getChildAffinityTargets(nodes, sys, CLASS_ENC, TYPE_NODE);
-    assert(!nodes.empty());
-    TargetHandle_t node = nodes[0];
-
-
     //Backplane Callout Sensor ID
     SensorID1 = UTIL::getSensorNumber(node, SENSOR_NAME_BACKPLANE_FAULT);
     memcpy(&o_data[index], &SensorID1, 4);
diff --git a/src/usr/htmgt/htmgt_occ.C b/src/usr/htmgt/htmgt_occ.C
index f7ae48a78..686a8338b 100644
--- a/src/usr/htmgt/htmgt_occ.C
+++ b/src/usr/htmgt/htmgt_occ.C
@@ -1022,6 +1022,51 @@ namespace HTMGT
                               HTMGT_RC_OCC_CRIT_FAILURE,
                               0, cv_safeReturnCode, 0, cv_safeOccInstance,
                               ERRORLOG::ERRL_SEV_UNRECOVERABLE);
+
+                    // Check if OCC already logged reason for safe mode
+                    // (add proc callout if non-OCC safe mode reason or
+                    //  the OCC hit an exception)
+                    TMGT_ERR("_resetOccs: Safe Mode (RC: 0x%04X OCC%d)",
+                             cv_safeReturnCode, cv_safeOccInstance);
+                    if (((cv_safeReturnCode & OCCC_COMP_ID) != OCCC_COMP_ID) ||
+                        ((cv_safeReturnCode & 0xE0) == 0xE0))
+                    {
+                        // Add processor callout
+                        TARGETING::ConstTargetHandle_t procTarget = nullptr;
+                        Occ *occPtr = _getOcc(cv_safeOccInstance);
+                        if (occPtr != nullptr)
+                        {
+                            procTarget =
+                                TARGETING::getParentChip(occPtr->getTarget());
+                        }
+                        else
+                        {
+                            TMGT_ERR("_resetOCCs: Unable to determine target, "
+                                     "using first proc");
+                            // Get all functional processors
+                            TARGETING::TargetHandleList pProcs;
+                            TARGETING::getChipResources(pProcs,
+                                                        TARGETING::TYPE_PROC,
+                                             TARGETING::UTIL_FILTER_FUNCTIONAL);
+                            procTarget = pProcs[0];
+                        }
+                        if (nullptr != procTarget)
+                        {
+                            const unsigned long huid =
+                                procTarget->getAttr<TARGETING::ATTR_HUID>();
+                            TMGT_ERR("_resetOCCs: Adding processor callout "
+                                     "(HUID=0x%0lX)", huid);
+                            err->addHwCallout(procTarget,
+                                              HWAS::SRCI_PRIORITY_MED,
+                                              HWAS::NO_DECONFIG,
+                                              HWAS::GARD_NULL);
+                        }
+                    }
+                    else
+                    {
+                        TMGT_INF("_resetOccs: OCC should have already logged "
+                                 "an error for safe mode");
+                    }
                 }
 
                 // Any error at this point means OCCs were not reactivated
@@ -1302,12 +1347,13 @@ namespace HTMGT
 
         // First add HTMGT specific data
         o_data[index++] = _getNumOccs();
-        o_data[index++] = (nullptr!=iv_occMaster)?iv_occMaster->getInstance():0xFF;
+        o_data[index++] =
+            (nullptr!=iv_occMaster)?iv_occMaster->getInstance():0xFF;
         o_data[index++] = iv_state;
         o_data[index++] = iv_targetState;
         o_data[index++] = iv_sysResetCount;
         o_data[index++] = iv_normalPstateTables ? 0 : 1;
-        index += 1; // reserved for expansion
+        o_data[index++] = 0x00; // STATUS VERSION (for future expansion)
         o_data[index++] = safeMode;
         UINT32_PUT(&o_data[index], cv_safeReturnCode);
         index += 4;
@@ -1322,11 +1368,13 @@ namespace HTMGT
             o_data[index++] = occ->getRole();
             o_data[index++] = occ->iv_masterCapable;
             o_data[index++] = occ->iv_commEstablished;
-            index += 3; // reserved for expansion
+            o_data[index++] = 0; // reserved for expansion
+            o_data[index++] = 0; // reserved for expansion
+            o_data[index++] = 0; // reserved for expansion
             o_data[index++] = occ->iv_failed;
             o_data[index++] = occ->needsReset();
             o_data[index++] = occ->iv_resetReason;
-            o_data[index++] = occ->iv_resetCount;
+            o_data[index++] = (occ->iv_wofResetCount<<4)|occ->iv_resetCount;
             if (occ->iv_lastPollValid)
             {
                 memcpy(&o_data[index], occ->iv_lastPollResponse, 4);
diff --git a/src/usr/htmgt/htmgt_utility.H b/src/usr/htmgt/htmgt_utility.H
index 5ef5d7457..0ffc63e7f 100644
--- a/src/usr/htmgt/htmgt_utility.H
+++ b/src/usr/htmgt/htmgt_utility.H
@@ -57,6 +57,20 @@
 #define TMGT_CONSOLE(_fmt_, _args_...)
 #endif
 
+
+// Enable this to force all HTMGT traces to console (for debug)
+#if 0
+#if defined(CONFIG_CONSOLE) && !defined(__HOSTBOOT_RUNTIME)
+#undef TMGT_INF
+#define    TMGT_INF( _fmt_, _args_...) \
+               CONSOLE::displayf( HTMGT_COMP_NAME, _fmt_, ##_args_ )
+#undef TMGT_ERR
+#define    TMGT_ERR( _fmt_, _args_...) \
+               CONSOLE::displayf( HTMGT_COMP_NAME, "E>" _fmt_, ##_args_ )
+#endif
+#endif
+
+
 inline uint16_t UINT16_GET(const uint8_t * i_ptr)
 {
     return (*i_ptr) << 8 | *(i_ptr+1);