DIMM Plugging Errors

This adds logic to change memory sensors to the sensor-specific offset
'Configuration error' whenever a memory plugging error appears in
an error log.

Change-Id: I94a58d55b8c81d106f77b00ee37c4d70a12a9fca
RTC:130381
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/20409
Tested-by: Jenkins Server
Tested-by: Jenkins OP Build CI
Tested-by: Jenkins OP HW
Tested-by: FSP CI Jenkins
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

1 files changed, 96 insertions, 54 deletions

diff --git a/src/usr/errl/errlmanager_common.C b/src/usr/errl/errlmanager_common.C
index 51ed8fa77..0020ee14e 100644
--- a/src/usr/errl/errlmanager_common.C
+++ b/src/usr/errl/errlmanager_common.C
@@ -449,6 +449,49 @@ void getSensorInfo(HWAS::callout_ud_t *i_ud,
                    uint8_t &o_eventOffset,
                    errlHndl_t& io_error );
 
+//  SensorModifier is a class that detects certain procedure callouts that
+//  can override the sensors generated by hardware callouts.
+//  For example, a memory plugging error can override a hardware callout
+//  of type memory in order to indicate a "configuration error" occured
+//  See errlmanager.H for more info
+
+// The constructor initializes flags for procedure callouts.
+inline SensorModifier::SensorModifier()
+{
+    iv_flag = 0;
+};
+
+// Consider this callout as a potential sensor modifier
+inline void SensorModifier::considerCallout(HWAS::callout_ud_t *i_ud)
+{
+    if (i_ud->type == HWAS::PROCEDURE_CALLOUT &&
+        i_ud->procedure == HWAS::EPUB_PRC_MEMORY_PLUGGING_ERROR)
+    {
+        iv_flag |= memory_plugging_error_mask;
+    }
+};
+
+// Modify the sensor if flag was set for that sensor.
+inline bool SensorModifier::modifySensor(uint8_t i_sensorType,
+        uint8_t& o_eventDirType, uint8_t& o_specificOffset)
+{
+    bool l_retval = false;
+    if( (iv_flag & memory_plugging_error_mask) &&
+       // we had a memory configuration error
+       i_sensorType == TARGETING::SENSOR_TYPE_MEMORY
+    )
+    {
+        o_eventDirType = IPMISEL::sensor_specific;
+                            // 0x6f Sensor-specific Offset
+        o_specificOffset = IPMISEL::event_data1_trans_monitor;
+                            // 0x07 Configuration Error
+        // modified
+        l_retval = true;
+    }
+    return l_retval;
+}
+
+
 ///////////////////////////////////////////////////////////////////////////////
 // ErrlManager::sendErrLogToBmc()
 ///////////////////////////////////////////////////////////////////////////////
@@ -459,6 +502,8 @@ void ErrlManager::sendErrLogToBmc(errlHndl_t &io_err, bool i_sendSels)
                 io_err->eid(), i_sendSels);
 
     do {
+        // keep track of procedure callouts that modify hardware callouts
+        SensorModifier l_modifier;
 
         // Decide whether we want to skip the error log
         if( io_err->getSkipShowingLog() )
@@ -469,13 +514,15 @@ void ErrlManager::sendErrLogToBmc(errlHndl_t &io_err, bool i_sendSels)
             break;
         }
 
-        std::vector<std::pair<uint8_t, uint8_t> > l_sensorNumbers;
+        // two pass algorithm to find the highest priority callout
+        // then send SELs for only the highest priority callouts
+        std::vector< HWAS::callout_ud_t* > l_callouts;
         HWAS::callOutPriority l_priority = HWAS::SRCI_PRIORITY_NONE;
         if (i_sendSels)
         {
             // look thru the errlog for any Callout UserDetail sections
-            // to determine the sensor information for the SEL
-            // create a vector of sensor numbers and offsets
+            // to later determine the sensor information for each SEL
+            // create a vector of callouts during the first pass
             for(std::vector<ErrlUD*>::const_iterator
                     it = io_err->iv_SectionVector.begin();
                     it != io_err->iv_SectionVector.end();
@@ -489,43 +536,21 @@ void ErrlManager::sendErrLogToBmc(errlHndl_t &io_err, bool i_sendSels)
                     HWAS::callout_ud_t *l_ud =
                         reinterpret_cast<HWAS::callout_ud_t*>((*it)->iv_pData);
 
-                    // if this callout isn't higher than any previous callout
-                    if (l_ud->priority < l_priority)
-                    {
-                        continue; // on to the next
-                    }
-
-                    // if greater, than clear out the previous list
+                    // if this callout is higher than any previous callout
                     if (l_ud->priority > l_priority)
                     {
                         TRACFCOMP(g_trac_errl,
                             "sendErrLogToBmc new priority picked 0x%x > 0x%x",
                             l_ud->priority, l_priority );
 
-                        //remove previous sensor data
-                        l_sensorNumbers.clear();
-
                         // and update the priority
                         l_priority = l_ud->priority;
                     }
+                    // consider as a potential modifier of other callouts
+                    l_modifier.considerCallout(l_ud);
 
-                    // greater than or equal - save this sensor
-
-                    //get the sensor number for the target
-                    uint8_t l_sensorNumber =
-                                TARGETING::UTIL::INVALID_IPMI_SENSOR;
-                    uint8_t l_eventOffset  =
-                                IPMISEL::event_data1_invalid_offset;
-
-                    getSensorInfo( l_ud, l_sensorNumber, l_eventOffset,
-                                      io_err);
-
-                    TRACFCOMP(g_trac_errl,
-                            "l_sensorNumber = 0x%x, l_eventOffset = 0x%x",
-                            l_sensorNumber, l_eventOffset );
-
-                    l_sensorNumbers.push_back(
-                            std::make_pair(l_sensorNumber, l_eventOffset));
+                    // add to list to be traversed in second pass
+                    l_callouts.push_back(l_ud);
                 } // if callout
             } // for each SectionVector
         } // if i_sendSels
@@ -555,49 +580,61 @@ void ErrlManager::sendErrLogToBmc(errlHndl_t &io_err, bool i_sendSels)
 
         if (i_sendSels)
         {
-            for(size_t i = 0; i < l_sensorNumbers.size(); i++)
+            for(size_t i = 0; i < l_callouts.size(); i++)
             {
-                uint8_t l_eventDirType  = IPMISEL::sensor_specific;
+                uint8_t l_eventDirType = IPMISEL::sensor_specific;
+                uint8_t l_sensorNumber = TARGETING::UTIL::INVALID_IPMI_SENSOR;
+                uint8_t l_eventOffset = IPMISEL::event_data1_invalid_offset;
+
+                // simplify callout notation
+                HWAS::callout_ud_t* l_ud = l_callouts.at(i);
+
+                // populate l_sensorNumber and l_eventOffset with values
+                getSensorInfo( l_ud, l_sensorNumber, l_eventOffset, io_err);
 
-                // if the offset is unknown after this then it will
-                // be updated based on elog severity below
-                uint8_t l_eventOffset = l_sensorNumbers.at(i).second ;
+                // if the offset is unknown at this point then it will
+                // be updated below by getSensorOffsetBasedOnSeverity
 
                 // last ditch effort, if no sensor number is present at this
                 // point, just use the system event sensor
-                if( l_sensorNumbers.at(i).first ==
+                if( l_sensorNumber ==
                         TARGETING::UTIL::INVALID_IPMI_SENSOR )
                 {
-                    l_sensorNumbers.at(i).first =
+                    l_sensorNumber =
                         TARGETING::UTIL::getSensorNumber(NULL,
                                 TARGETING::SENSOR_NAME_SYSTEM_EVENT);
 
-                    l_sensorNumbers.at(i).second =
+                    l_eventOffset =
                         SENSOR::UNDETERMINED_SYSTEM_HW_FAILURE;
 
                 }
 
                 // grab the sensor type so the bmc knows how to use the offset
                 uint8_t unused = 0;
-                uint8_t l_SensorType = 0;
+                uint8_t l_sensorType = 0;
 
                 errlHndl_t e =
                     SENSOR::SensorBase::getSensorType(
-                            l_sensorNumbers.at(i).first,
-                            l_SensorType,unused);
+                            l_sensorNumber,
+                            l_sensorType,unused);
 
                 if( e )
                 {
                     TRACFCOMP(g_trac_errl,
                             ERR_MRK"Failed to get sensor type for sensor %d",
-                            l_sensorNumbers.at(i).first);
+                            l_sensorNumber);
 
-                    l_SensorType = 0;
+                    l_sensorType = 0;
                     // since we are in the commit path, lets just delete this
                     // error and move on.
                     delete e;
                 }
 
+                // this call will modify the sensor if any procedure callout is
+                // known to change its effect
+                bool l_wasModified = l_modifier.modifySensor(l_sensorType,
+                                                l_eventDirType, l_eventOffset);
+
                 // if no offset has been configured set it based on the severity
                 if( l_eventOffset == IPMISEL::event_data1_invalid_offset )
                 {
@@ -609,17 +646,22 @@ void ErrlManager::sendErrLogToBmc(errlHndl_t &io_err, bool i_sendSels)
                 // pel data, otherwise we send no data
                 uint32_t selSize = ( i == 0 ) ? l_pelSize:0;
 
-                TRACFCOMP(g_trac_errl, INFO_MRK
-                        "sendErrLogToBmc: sensor %.2x/%.2x event %x/%x, size %d",
-                        l_SensorType, l_sensorNumbers.at(i).first,
-                        l_eventDirType, l_eventOffset, selSize );
-
-                IPMISEL::sendESEL(l_pelData, selSize,
-                                io_err->eid(),
-                                l_eventDirType, l_eventOffset,
-                                l_SensorType,
-                                l_sensorNumbers.at(i).first);
-            } // for l_sensorNumbers
+                // only send highest priority SELs or
+                // SELs of lesser priority that were modified
+                if (l_ud->priority == l_priority || l_wasModified)
+                {
+                    TRACFCOMP(g_trac_errl, INFO_MRK "sendErrLogToBmc:"
+                            " sensor %.2x/%.2x event %x/%x, size %d",
+                            l_sensorType, l_sensorNumber,
+                            l_eventDirType, l_eventOffset, selSize );
+
+                    IPMISEL::sendESEL(l_pelData, selSize,
+                                    io_err->eid(),
+                                    l_eventDirType, l_eventOffset,
+                                    l_sensorType,
+                                    l_sensorNumber);
+                }
+            } // for l_callouts
         }
         else
         {