PRD: initializing MemTdCtlr variables for broadcast mode

Change-Id: I4982ef4c27f694e35471cad3e3ec57eedfa2632e
CQ: SW418155
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/54548
Tested-by: Jenkins Server <pfd-jenkins+hostboot@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>
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@us.ibm.com>
Reviewed-by: Brian J. Stegmiller <bjs@us.ibm.com>
Reviewed-by: Caleb N. Palmer <cnpalmer@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>

4 files changed, 101 insertions, 111 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.C
index c78e71571..ba1eadd83 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.C
@@ -86,8 +86,6 @@ uint32_t MemTdCtlr<T>::handleCmdComplete( STEP_CODE_DATA_STRUCT & io_sc )
 
     do
     {
-        #ifdef __HOSTBOOT_RUNTIME
-
         // Make sure the TD controller is initialized.
         o_rc = initialize();
         if ( SUCCESS != o_rc )
@@ -96,7 +94,7 @@ uint32_t MemTdCtlr<T>::handleCmdComplete( STEP_CODE_DATA_STRUCT & io_sc )
             break;
         }
 
-        #else // IPL only
+        #ifndef __HOSTBOOT_RUNTIME // IPL only
 
         // TODO: RTC 179251 asserting here doesn't give us enough FFDC to debug
         //       why we got this erroneous attention. Eventually, we will want
@@ -142,36 +140,6 @@ uint32_t MemTdCtlr<T>::handleCmdComplete( STEP_CODE_DATA_STRUCT & io_sc )
             // informational error logs.
             if ( !errorsFound ) io_sc.service_data->setDontCommitErrl();
         }
-        else
-        {
-            // Make sure iv_stoppedRank still gets updated.
-            std::vector<ExtensibleChip *> portList;
-            o_rc = getMcbistMaintPort( iv_chip, portList );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "getMcbistMaintPort(0x%08x) failed",
-                        iv_chip->getHuid() );
-                break;
-            }
-
-            // In broadcast mode, the rank configuration for all ports will be
-            // the same. In non-broadcast mode, there will only be one MCA in
-            // the list. Therefore, we can simply use the first MCA in the list
-            // for all configs.
-            ExtensibleChip * stopChip = portList.front();
-
-            // Get the address in which the command stopped.
-            MemAddr addr;
-            o_rc = getMemMaintAddr<T>( iv_chip, addr );
-            if ( SUCCESS != o_rc )
-            {
-                PRDF_ERR( PRDF_FUNC "getMemMaintAddr<T>(0x%08x) failed",
-                          iv_chip->getHuid() );
-                break;
-            }
-
-            iv_stoppedRank = __getStopRank<TYPE_MCA>( stopChip, addr );
-        }
 
         // Move onto the next step in the state machine.
         o_rc = nextStep( io_sc );
@@ -188,7 +156,10 @@ uint32_t MemTdCtlr<T>::handleCmdComplete( STEP_CODE_DATA_STRUCT & io_sc )
     //       successfully with no errors because the error log will not be
     //       committed.
     if ( !io_sc.service_data->queryDontCommitErrl() )
+    {
+        collectStateCaptureData( io_sc, TD_CTLR_DATA::END );
         MemCaptureData::addEccData<T>( iv_chip, io_sc );
+    }
 
     if ( SUCCESS != o_rc )
     {
@@ -219,10 +190,6 @@ uint32_t MemTdCtlr<T>::handleCmdComplete( STEP_CODE_DATA_STRUCT & io_sc )
 
         #endif
     }
-    else
-    {
-        collectStateCaptureData( io_sc, TD_CTLR_DATA::END );
-    }
 
     return o_rc;
 
@@ -249,9 +216,6 @@ template<TARGETING::TYPE T>
 uint32_t __analyzeCmdComplete( ExtensibleChip * i_chip,
                                TdQueue & io_queue,
                                TdRankListEntry & o_stoppedRank,
-                               #ifndef __HOSTBOOT_RUNTIME
-                               bool & o_broadcastMode,
-                               #endif
                                const MemAddr & i_addr,
                                bool & o_errorsFound,
                                STEP_CODE_DATA_STRUCT & io_sc );
@@ -260,9 +224,6 @@ template<>
 uint32_t __analyzeCmdComplete<TYPE_MCBIST>( ExtensibleChip * i_chip,
                                             TdQueue & io_queue,
                                             TdRankListEntry & o_stoppedRank,
-                                            #ifndef __HOSTBOOT_RUNTIME
-                                            bool & o_broadcastMode,
-                                            #endif
                                             const MemAddr & i_addr,
                                             bool & o_errorsFound,
                                             STEP_CODE_DATA_STRUCT & io_sc )
@@ -294,11 +255,6 @@ uint32_t __analyzeCmdComplete<TYPE_MCBIST>( ExtensibleChip * i_chip,
         // Update iv_stoppedRank.
         o_stoppedRank = __getStopRank<TYPE_MCA>( stopChip, i_addr );
 
-        #ifndef __HOSTBOOT_RUNTIME
-        // Update iv_broadcastMode.
-        o_broadcastMode = ( 1 < portList.size() );
-        #endif
-
         // Check each MCA for ECC errors.
         for ( auto & mcaChip : portList )
         {
@@ -330,9 +286,6 @@ template<>
 uint32_t __analyzeCmdComplete<TYPE_MBA>( ExtensibleChip * i_chip,
                                          TdQueue & io_queue,
                                          TdRankListEntry & o_stoppedRank,
-                                         #ifndef __HOSTBOOT_RUNTIME
-                                         bool & o_broadcastMode,
-                                         #endif
                                          const MemAddr & i_addr,
                                          bool & o_errorsFound,
                                          STEP_CODE_DATA_STRUCT & io_sc )
@@ -371,9 +324,6 @@ uint32_t MemTdCtlr<T>::analyzeCmdComplete( bool & o_errorsFound,
 
         // Then, check for ECC errors, if they exist.
         o_rc = __analyzeCmdComplete<T>( iv_chip, iv_queue, iv_stoppedRank,
-                                        #ifndef __HOSTBOOT_RUNTIME
-                                        iv_broadcastMode,
-                                        #endif
                                         addr, o_errorsFound, io_sc );
         if ( SUCCESS != o_rc )
         {
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H
index 8c0ea9df2..3f3f354fc 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
+/* Contributors Listed Below - COPYRIGHT 2016,2018                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -57,25 +57,23 @@ class MemTdCtlr
      * This contructor will only be called in the MCBIST or MBA data bundle,
      * which already checks for a valid type.
      *
+     * Need to initialize iv_stoppedRank to a valid entry in iv_rankList. Use
+     * the last entry in the list so that the 'next' rank is the first entry
+     * in the list.
+     *
      * @param i_chip An MCBIST or MBA chip.
      */
     explicit MemTdCtlr( ExtensibleChip * i_chip ) :
         iv_chip( i_chip ), iv_rankList( i_chip ),
-        iv_stoppedRank( i_chip, MemRank(0) )
+        iv_stoppedRank( iv_rankList.getList().back() )
     {
         PRDF_ASSERT( T == iv_chip->getType() );
-        if ( TARGETING::TYPE_MCBIST == i_chip->getType() )
-        {
-            ExtensibleChip * mcaChip =
-                PlatServices::getConnectedChild(i_chip, TARGETING::TYPE_MCA, 0);
-            iv_stoppedRank = TdRankListEntry( mcaChip, MemRank(0) );
-        }
     }
 
     /**
      * @brief  Determines and executes the next course of action after a
      *         maintenance command complete attention.
-     * @note   Initializes the TD controller, if needed (runtime only).
+     * @note   Initializes the TD controller, if needed.
      * @param  io_sc The step code data struct.
      * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
      */
@@ -97,8 +95,8 @@ class MemTdCtlr
      * requests will be ignored. Any chip marks placed during this time will be
      * redetected when the runtime TD controller is initialized.
      *
-     * During IPL, this will simply add a new procedure to the queue, since we
-     * know TD will already be in progress when this is called.
+     * During MemDiags, this will simply add a new procedure to the queue, since
+     * we know a TD procedure will already be in progress when this is called.
      *
      * @note   Initializes the TD controller, if needed.
      * @param  io_sc    The step code data struct.
@@ -118,14 +116,23 @@ class MemTdCtlr
 
     /**
      * @brief Handles reset-reload or FO scenario.
-     * @note  This function will check if PRD was unable to restart maintenance
-     *        command before R/R or FO. In that scenario, this function will
-     *        start maintenance command. As during R/R or F/O we do not have any
-     *        mechanism to restore the complete state of TD controller, we will
-     *        not start any interrupted or pending TD procedure. We will only
-     *        start BG scrub. If we found any chip marks during TD state
-     *        machine initialize we will start VCM procedure rather than
-     *        BG scrub.
+     *
+     * This does not call initialize() or start any maintenance commands.
+     * Instead, it checks the hardware's current state and ensures by the end of
+     * the function that either a command is currently running or there will be
+     * a command complete attention pending that PRD will handle separately.
+     *
+     * If there is already an active command complete attention. This function
+     * does nothing because PRD will handle the attention soon.
+     *
+     * If there is no active command complete attention and there is no command
+     * currently in progress, it will set the command complete attention and PRD
+     * will handle that attention soon.
+     *
+     * Otherwise, there is a command in progress. So, it will check for any
+     * unverified chip marks. If any exist, it will force the current command to
+     * stop, causing a command complete attention that PRD will handle soon.
+     *
      * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
      */
     uint32_t handleRrFo();
@@ -135,6 +142,24 @@ class MemTdCtlr
   private:
 
     /**
+     * @brief  Initializes the TD controller, if needed.
+     *
+     * This should be called at the beginning of every public function to ensure
+     * the TD controller is initialized.
+     *
+     * During MemDiags, this initializes iv_broadcastModeCapable.
+     *
+     * At runtime, this is used to query hardware for any unverified chip marks
+     * that may have occurred after starting background scrubbing, but before
+     * PRD is up and running. We may also have unverified chip marks if the HBRT
+     * service is stopped and restarted (PRD is reinitialize and all previous
+     * state machine data is lost).
+     *
+     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+     */
+    uint32_t initialize();
+
+    /**
      * @brief This is called when there are no more TD procedures to execute.
      *
      * During Memory Diagnostics, this means the current pattern test command
@@ -198,8 +223,8 @@ class MemTdCtlr
 
     /**
      * @brief  This is called when handling a command complete attention for a
-     *         non-TD command to initialize iv_stoppedRank and iv_broadcastMode
-     *         then check for any ECC errors.
+     *         non-TD command to initialize iv_stoppedRank then check for any
+     *         ECC errors.
      * @param  o_errorsFound True if errors where found and handled. False
      *                       otherwise.
      * @param  io_sc         The step code data struct.
@@ -237,22 +262,6 @@ class MemTdCtlr
      */
     uint32_t unmaskEccAttns();
 
-    /**
-     * @brief  Initializes the TD controller, if needed.
-     *
-     * This is only supported during runtime. This is mostly useful at runtime
-     * to query hardware for any unverified chip marks. Those may occur after
-     * starting background scrubbing, but before PRD is up and running. We may
-     * also have unverified chip marks if the HBRT service is stopped and
-     * restarted (PRD is reinitialize and all previous state machine data is
-     * lost).
-     *
-     * @note   Should be called at the beginning of every public function to
-     *         ensure the TD controller is initialized.
-     * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
-     */
-    uint32_t initialize();
-
     #endif
 
   private: // instance variables
@@ -276,11 +285,11 @@ class MemTdCtlr
      *  diagnostics are complete. */
     TdRankListEntry iv_stoppedRank;
 
-    #ifdef __HOSTBOOT_RUNTIME
-
     /** True if the TD controller has been initialized. False otherwise. */
     bool iv_initialized = false;
 
+    #ifdef __HOSTBOOT_RUNTIME
+
     /** True if background scrubbing should be resumed after pausing on error.
      *  False if a TD procedure had been executed and background scrubbing needs
      *  to be restarted with a new command. */
@@ -288,9 +297,8 @@ class MemTdCtlr
 
     #else // IPL only
 
-    /** Combined with iv_stoppedRank. Indicates if the non-TD command that
-     *  stopped was in broadcast mode or not. */
-    bool iv_broadcastMode = false;
+    /** Indicates if broadcast mode is capable on iv_chip. */
+    bool iv_broadcastModeCapable = false;
 
     #endif
 
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
index e6d769b6c..7caba811b 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_ipl.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016,2017                        */
+/* Contributors Listed Below - COPYRIGHT 2016,2018                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -65,6 +65,32 @@ uint32_t MemTdCtlr<T>::handleTdEvent( STEP_CODE_DATA_STRUCT & io_sc,
 //------------------------------------------------------------------------------
 
 template <TARGETING::TYPE T>
+uint32_t MemTdCtlr<T>::initialize()
+{
+    #define PRDF_FUNC "[MemTdCtlr::initialize] "
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        if ( iv_initialized ) break; // nothing to do
+
+        // Check if broadcast mode is capable on this chip.
+        iv_broadcastModeCapable = isBroadcastModeCapable<T>( iv_chip );
+
+        // At this point, the TD controller is initialized.
+        iv_initialized = true;
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
+template <TARGETING::TYPE T>
 uint32_t MemTdCtlr<T>::defaultStep( STEP_CODE_DATA_STRUCT & io_sc )
 {
     #define PRDF_FUNC "[MemTdCtlr::defaultStep] "
@@ -72,7 +98,7 @@ uint32_t MemTdCtlr<T>::defaultStep( STEP_CODE_DATA_STRUCT & io_sc )
     uint32_t o_rc = SUCCESS;
 
     TdRankListEntry nextRank = iv_rankList.getNext( iv_stoppedRank,
-                                                    iv_broadcastMode );
+                                                    iv_broadcastModeCapable );
 
     do
     {
diff --git a/src/usr/diag/prdf/plat/mem/prdfRestoreDramRepairs.C b/src/usr/diag/prdf/plat/mem/prdfRestoreDramRepairs.C
index 3c34320b3..1fc61fba6 100644
--- a/src/usr/diag/prdf/plat/mem/prdfRestoreDramRepairs.C
+++ b/src/usr/diag/prdf/plat/mem/prdfRestoreDramRepairs.C
@@ -54,6 +54,7 @@ namespace RDR // local utility functions to support PRDF::restoreDramRepairs()
 {
 
 // Creates and returns an error log.
+template<TARGETING::TYPE T>
 errlHndl_t createErrl( uint32_t i_reasonCode, TargetHandle_t i_trgt,
                        uint32_t i_signature )
 {
@@ -63,12 +64,18 @@ errlHndl_t createErrl( uint32_t i_reasonCode, TargetHandle_t i_trgt,
     // Note that the error log tags are not needed because PRD uses its own
     // signature parser.
 
-    return new ERRORLOG::ErrlEntry(
+    errlHndl_t errl = new ERRORLOG::ErrlEntry(
                             ERRORLOG::ERRL_SEV_PREDICTIVE,  // severity
                             PRDF_RESTORE_DRAM_REPAIR,       // module ID
                             i_reasonCode,                   // reason code
                             userdata12,                     // user data 1 & 2
                             userdata34 );                   // user data 3 & 4
+
+    // Add capture data. Need to do this now before the DIMM callouts are made
+    // because the VPD is cleared if a DIMM is added to the callout list.
+    MemCaptureData::addEccData<T>( i_trgt, errl );
+
+    return errl;
 }
 
 //------------------------------------------------------------------------------
@@ -80,9 +87,6 @@ void commitErrl( errlHndl_t i_errl, TargetHandle_t i_trgt )
 {
     if ( NULL != i_errl )
     {
-        // Add capture data
-        MemCaptureData::addEccData<T>( i_trgt, i_errl );
-
         // Add traces
         i_errl->collectTrace( PRDF_COMP_NAME, 512 );
 
@@ -147,7 +151,7 @@ void commitSoftError( uint32_t i_reasonCode, TargetHandle_t i_trgt,
 {
     if ( i_analysisErrors )
     {
-        errlHndl_t errl = createErrl( i_reasonCode, i_trgt, i_signature );
+        errlHndl_t errl = createErrl<T>( i_reasonCode, i_trgt, i_signature );
         errl->addProcedureCallout( HWAS::EPUB_PRC_LVL_SUPP,
                                    HWAS::SRCI_PRIORITY_HIGH );
         commitErrl<T>( errl, i_trgt );
@@ -225,8 +229,9 @@ bool processRepairedRanks<TYPE_MCA>( TargetHandle_t i_trgt,
 
                 if ( NULL == errl )
                 {
-                    errl = createErrl( PRDF_DETECTED_FAIL_HARDWARE, i_trgt,
-                                       PRDFSIG_RdrRepairsUsed );
+                    errl = createErrl<TYPE_MCA>( PRDF_DETECTED_FAIL_HARDWARE,
+                                                 i_trgt,
+                                                 PRDFSIG_RdrRepairsUsed );
                 }
 
                 std::vector<MemSymbol> symList;
@@ -344,8 +349,9 @@ bool processRepairedRanks<TYPE_MBA>( TargetHandle_t i_trgt,
 
                 if ( NULL == errl )
                 {
-                    errl = createErrl( PRDF_DETECTED_FAIL_HARDWARE, i_mba,
-                                       PRDFSIG_RdrRepairsUsed );
+                    errl = createErrl<TYPE_MBA>( PRDF_DETECTED_FAIL_HARDWARE,
+                                                 i_mba,
+                                                 PRDFSIG_RdrRepairsUsed );
                 }
 
                 std::vector<CenSymbol> list;
@@ -425,8 +431,8 @@ bool processBadDimms<TYPE_MCA>( TargetHandle_t i_trgt, uint8_t i_badDimmMask )
         {
             if ( NULL == errl )
             {
-                errl = createErrl( PRDF_DETECTED_FAIL_HARDWARE, i_trgt,
-                                   PRDFSIG_RdrRepairUnavail );
+                errl = createErrl<TYPE_MCA>( PRDF_DETECTED_FAIL_HARDWARE,
+                                             i_trgt, PRDFSIG_RdrRepairUnavail );
             }
 
             __calloutDimm<TYPE_MCA, DIMMS_PER_RANK::MCA>( errl, i_trgt, dimm );
@@ -492,8 +498,8 @@ bool processBadDimms<TYPE_MBA>( TargetHandle_t i_trgt, uint8_t i_badDimmMask )
         {
             if ( NULL == errl )
             {
-                errl = createErrl( PRDF_DETECTED_FAIL_HARDWARE, i_mba,
-                                   PRDFSIG_RdrRepairUnavail );
+                errl = createErrl<TYPE_MBA>( PRDF_DETECTED_FAIL_HARDWARE, i_mba,
+                                             PRDFSIG_RdrRepairUnavail );
             }
 
             o_calloutMade = true;