PRD: refined handleChnlFail() for Centaur

Change-Id: I698b0320a51cb9452a6b0e661c6f3c24a8f70b78
RTC: 136123
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/58607
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/59762
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@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>

src/usr/diag/prdf/common/plat/cen/prdfCenMembuf_common.C

@@ -63,6 +63,23 @@ int32_t Initialize( ExtensibleChip * i_chip )
 }
 PRDF_PLUGIN_DEFINE( cen_centaur, Initialize );
 
+/**
+ * @brief  Analysis code that is called before the main analyze() function.
+ * @param  i_chip     A MEMBUF chip.
+ * @param  io_sc      The step code data struct.
+ * @param  o_analyzed True if analysis is done on this chip, false otherwise.
+ * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+ */
+int32_t PreAnalysis( ExtensibleChip * i_chip, STEP_CODE_DATA_STRUCT & io_sc,
+                     bool & o_analyzed )
+{
+    o_analyzed = false;
+
+    // Check for a channel failure before analyzing this chip.
+    return MemUtils::handleChnlFail<TYPE_MEMBUF>( i_chip, io_sc );
+}
+PRDF_PLUGIN_DEFINE( cen_centaur, PreAnalysis );
+
 /**
  * @brief  Plugin function called after analysis is complete but before PRD
  *         exits.
@@ -108,11 +125,15 @@ int32_t analyzeDmiChnlFail( ExtensibleChip * i_mbChip,
 
     do
     {
+        // Query the connected DMI for channel fail attentions.
         ExtensibleChip * dmiChip = getConnectedParent( i_mbChip, TYPE_DMI );
-
-        // TODO: RTC 136123 Need to call new interface that queries if there was
-        //       a channel fail attention on the other side of the interface.
         bool dmiChnlFail = false;
+        if ( SUCCESS != queryChnlFail<TYPE_DMI>(dmiChip, dmiChnlFail) )
+        {
+            PRDF_ERR( PRDF_FUNC "queryChnlFail(0x%08x) failed",
+                      dmiChip->getHuid() );
+            break;
+        }
 
         // If there is a channel fail attention on the other side of the bus,
         // analyze the DMI target.

src/usr/diag/prdf/common/plat/mem/prdfMemUtils.C

@@ -506,69 +506,153 @@ void cleanupChnlAttns<TYPE_MEMBUF>( ExtensibleChip * i_chip,
 
 //------------------------------------------------------------------------------
 
-/* TODO RTC 136123
-int32_t checkMcsChannelFail( ExtensibleChip * i_mcsChip,
-                             STEP_CODE_DATA_STRUCT & io_sc )
+template<TARGETING::TYPE T>
+uint32_t __queryChnlFail( ExtensibleChip * i_chip, bool & o_chnlFail );
+
+template<>
+uint32_t __queryChnlFail<TYPE_MEMBUF>( ExtensibleChip * i_chip,
+                                       bool & o_chnlFail )
 {
-    #define PRDF_FUNC "[MemUtils::checkMcsChannelFail] "
+    #define PRDF_FUNC "[MemUtils::__queryChnlFail] "
 
-    int32_t o_rc = SUCCESS;
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MEMBUF == i_chip->getType() );
 
-    do
-    {
-        // Skip if already handling unit checkstop.
-        if ( io_sc.service_data->IsUnitCS() )
-            break;
+    uint32_t o_rc = SUCCESS;
 
-        // Must be an MCS.
-        if ( TYPE_MCS != getTargetType(i_mcsChip->GetChipHandle()) )
-        {
-            PRDF_ERR( PRDF_FUNC "i_mcsChip is not TYPE_MCS" );
-            o_rc = FAIL; break;
-        }
+    o_chnlFail = false;
 
-        // Check MCIFIR[31] for presence of channel fail.
-        SCAN_COMM_REGISTER_CLASS * mcifir = i_mcsChip->getRegister("MCIFIR");
-        o_rc = mcifir->Read();
+    do
+    {
+        // Simply check the Centaur CS global reg for active attentions.
+        SCAN_COMM_REGISTER_CLASS * fir = i_chip->getRegister("GLOBAL_CS_FIR");
+        o_rc = fir->Read();
         if ( SUCCESS != o_rc )
         {
-            PRDF_ERR( PRDF_FUNC "Read() failed on MCIFIR" );
+            PRDF_ERR( PRDF_FUNC "Failed to read GLOBAL_CS_FIR on 0x%08x",
+                      i_chip->getHuid() );
             break;
         }
 
-        if ( !mcifir->IsBitSet(31) ) break; // No channel fail, so exit.
+        o_chnlFail = !fir->BitStringIsZero();
 
-        // Set unit checkstop flag and cause attention type.
-        io_sc.service_data->setFlag(ServiceDataCollector::UNIT_CS);
-        io_sc.service_data->setSecondaryAttnType(UNIT_CS);
-        io_sc.service_data->SetThresholdMaskId(0);
+    } while (0);
 
-        // Set it as SUE generation point.
-        io_sc.service_data->SetUERE();
+    return o_rc;
 
-        // Indicate that cleanup is required.
-        P8McsDataBundle * mcsdb = getMcsDataBundle( i_mcsChip );
-        ExtensibleChip * membChip =  mcsdb->getMembChip();
-        if ( NULL == membChip )
-        {
-            PRDF_ERR( PRDF_FUNC "getMembChip() returned NULL" );
-            o_rc = FAIL; break;
-        }
-        MembufDataBundle * mbdb = getMembufDataBundle( membChip );
-        mbdb->iv_doChnlFailCleanup = true;
+    #undef PRDF_FUNC
+}
+
+template<>
+uint32_t __queryChnlFail<TYPE_DMI>( ExtensibleChip * i_chip, bool & o_chnlFail )
+{
+    // There is a HWP on the processor side that will query the CHIFIR, IOMCFIR,
+    // and associated configuration registers for a valid channel failure
+    // attention.
+    return PlatServices::queryChnlFail<TYPE_DMI>( i_chip, o_chnlFail );
+}
+
+//------------------------------------------------------------------------------
+
+template<TARGETING::TYPE T>
+void __setChnlFailCleanup( ExtensibleChip * i_chip );
+
+template<>
+void __setChnlFailCleanup<TYPE_MEMBUF>( ExtensibleChip * i_chip )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MEMBUF == i_chip->getType() );
+
+    getMembufDataBundle(i_chip)->iv_doChnlFailCleanup = true;
+}
+
+template<>
+void __setChnlFailCleanup<TYPE_DMI>( ExtensibleChip * i_chip )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_DMI == i_chip->getType() );
+
+    ExtensibleChip * membChip = getConnectedChild( i_chip, TYPE_MEMBUF, 0 );
+    PRDF_ASSERT( nullptr != membChip ); // shouldn't be possible
+
+    __setChnlFailCleanup<TYPE_MEMBUF>( membChip );
+}
+
+//------------------------------------------------------------------------------
+
+template<TARGETING::TYPE T>
+uint32_t handleChnlFail( ExtensibleChip * i_chip,
+                         STEP_CODE_DATA_STRUCT & io_sc )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( T == i_chip->getType() );
+
+    uint32_t o_rc = SUCCESS;
+
+    do
+    {
+        // Skip if already handling channel failure.
+        if ( io_sc.service_data->IsUnitCS() ) break;
+
+        // Skip if currently analyzing a host attention. This is a required for
+        // a rare scenario when a channel failure occurs after PRD is called to
+        // handle the host attention.
+        if ( HOST_ATTN == io_sc.service_data->getPrimaryAttnType() ) break;
+
+        // Look for the channel fail attention.
+        bool isChnlFail = false;
+        uint32_t o_rc = __queryChnlFail<T>( i_chip, isChnlFail );
+        if ( SUCCESS != o_rc ) break;
+
+        if ( ! isChnlFail ) break; // No channel fail, nothing more to do.
+
+        // Change the secondary attention type to UNIT_CS so the rule code will
+        // start looking for UNIT_CS attentions instead of recoverable.
+        io_sc.service_data->setSecondaryAttnType( UNIT_CS );
+
+        // Set the UNIT_CS flag in the SDC to indicate a channel failure has
+        // been detected and there is no need to check again.
+        io_sc.service_data->setFlag( ServiceDataCollector::UNIT_CS );
+
+        // Make the error log predictive and set threshold.
+        io_sc.service_data->setFlag( ServiceDataCollector::SERVICE_CALL );
+        io_sc.service_data->setFlag( ServiceDataCollector::AT_THRESHOLD );
+
+        // Channel failures will always send SUEs.
+        io_sc.service_data->setFlag( ServiceDataCollector::UERE );
+
+        // Indicate cleanup is required on this channel.
+        __setChnlFailCleanup<T>( i_chip );
 
     } while (0);
 
-    if ( SUCCESS != o_rc )
+    return o_rc;
+}
+
+template
+uint32_t handleChnlFail<TYPE_MEMBUF>( ExtensibleChip * i_chip,
+                                      STEP_CODE_DATA_STRUCT & io_sc );
+template
+uint32_t handleChnlFail<TYPE_DMI>( ExtensibleChip * i_chip,
+                                   STEP_CODE_DATA_STRUCT & io_sc );
+
+template<>
+uint32_t handleChnlFail<TYPE_MC>( ExtensibleChip * i_chip,
+                                  STEP_CODE_DATA_STRUCT & io_sc )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MC == i_chip->getType() );
+
+    uint32_t o_rc = SUCCESS;
+
+    for ( auto & dmiChip : getConnected(i_chip, TYPE_DMI) )
     {
-        PRDF_ERR( PRDF_FUNC "Failed: i_mcsChip=0x%08x", i_mcsChip->GetId() );
+        o_rc = handleChnlFail<TYPE_DMI>( dmiChip, io_sc );
+        if ( SUCCESS != o_rc ) break;
     }
 
     return o_rc;
-
-    #undef PRDF_FUNC
 }
-*/
 
 //------------------------------------------------------------------------------
 

src/usr/diag/prdf/common/plat/mem/prdfMemUtils.H

@@ -109,15 +109,6 @@ int32_t collectCeStats( ExtensibleChip * i_chip, const MemRank & i_rank,
 template<TARGETING::TYPE T>
 uint8_t getDramSize( ExtensibleChip * i_chip, uint8_t i_dimmSlct = 0 );
 
-/**
- * @brief  Check for channel fail attentions on the MCS side of the DMI bus.
- * @param  i_mcsChip An MCS chip.
- * @param  io_sc     The step code data struct.
- * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
- */
-int32_t checkMcsChannelFail( ExtensibleChip * i_mcsChip,
-                             STEP_CODE_DATA_STRUCT & io_sc );
-
 /**
  * @brief  determines the type of Centaur based raw card  associated with MBA.
  * @param  i_mba    mba target
@@ -140,6 +131,25 @@ int32_t getRawCardType( TARGETING::TargetHandle_t i_mba,
 template<TARGETING::TYPE T>
 void cleanupChnlAttns( ExtensibleChip * i_chip, STEP_CODE_DATA_STRUCT & io_sc );
 
+/**
+ * @brief  Checks for channel fail attentions on the target side of the bus.
+ * @note   If a channel fail attention is present, this function will set the
+ *         secondary attention type to UNIT_CS, among other appropriate flags
+ *         in the SDC. It is important that this is called in the PreAnalysis
+ *         plugin so that PRD knows to look for UNIT_CS attentions instead of
+ *         recoverable attentions.
+ * @note   This only checks one side of the bus. It does not check both sides.
+ * @note   If the secondary attention type is HOST_ATTN, the function exits and
+ *         waits for the next attention. This is for the rare scenario where a
+ *         channel failure occurs after PRD is called to handle a HOST_ATTN.
+ * @param  i_chip MEMBUF, DMI, or MC chip.
+ * @param  io_sc  The step code data struct.
+ * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+ */
+template<TARGETING::TYPE T>
+uint32_t handleChnlFail( ExtensibleChip * i_chip,
+                         STEP_CODE_DATA_STRUCT & io_sc );
+
 /**
  * @brief  After analyzing a memory channel failure, we want to mask off all
  *         possible attentions on that channel to ensure we don't get any errant

src/usr/diag/prdf/common/plat/p9/prdfP9Dmi_common.C

@@ -49,6 +49,23 @@ namespace p9_dmi
 //
 //##############################################################################
 
+/**
+ * @brief  Analysis code that is called before the main analyze() function.
+ * @param  i_chip     A DMI chip.
+ * @param  io_sc      The step code data struct.
+ * @param  o_analyzed True if analysis is done on this chip, false otherwise.
+ * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+ */
+int32_t PreAnalysis( ExtensibleChip * i_chip, STEP_CODE_DATA_STRUCT & io_sc,
+                     bool & o_analyzed )
+{
+    o_analyzed = false;
+
+    // Check for a channel failure before analyzing this chip.
+    return MemUtils::handleChnlFail<TYPE_DMI>( i_chip, io_sc );
+}
+PRDF_PLUGIN_DEFINE( p9_dmi, PreAnalysis );
+
 /**
  * @brief  Plugin function called after analysis is complete but before PRD
  *         exits.

src/usr/diag/prdf/common/plat/p9/prdfP9Mc_common.C

@@ -49,6 +49,23 @@ namespace p9_mc
 //
 //##############################################################################
 
+/**
+ * @brief  Analysis code that is called before the main analyze() function.
+ * @param  i_chip     An MC chip.
+ * @param  io_sc      The step code data struct.
+ * @param  o_analyzed True if analysis is done on this chip, false otherwise.
+ * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+ */
+int32_t PreAnalysis( ExtensibleChip * i_chip, STEP_CODE_DATA_STRUCT & io_sc,
+                     bool & o_analyzed )
+{
+    o_analyzed = false;
+
+    // Check for a channel failure before analyzing this chip.
+    return MemUtils::handleChnlFail<TYPE_MC>( i_chip, io_sc );
+}
+PRDF_PLUGIN_DEFINE( p9_mc, PreAnalysis );
+
 /**
  * @brief  Plugin function called after analysis is complete but before PRD
  *         exits.