PRD: Shut down mirror on threshold of mirrored memory UEs

Change-Id: I39c0d302788092378dea15dbf106e4e600ef6ae2
CQ: SW274471
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/13483
Reviewed-by: Zane Shelley <zshelle@us.ibm.com>
Tested-by: Jenkins Server
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/14600
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

4 files changed, 156 insertions, 7 deletions

diff --git a/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule b/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule
index 58c2f6ca2..008b4f1cb 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule
+++ b/src/usr/diag/prdf/common/plat/pegasus/Mcs.rule
@@ -154,7 +154,19 @@ chip Mcs
     {
         name        "MC0.MCS0.LEFT.LEFT.MCFGPQ";
         scomaddr    0x02011800;
-        capture     group never;
+        capture     group default;
+    };
+
+    ############################################################################
+    # MCS Hardware Force Mirror Read SUE
+    ############################################################################
+
+    register MCHWFM
+    {
+        name        "MC0.MCS0.LEFT.LEFT.MCHWFMQ";
+        scomaddr    0x0201181C;
+        capture     group MirrorRegs; # Only valid on even MCSs of the mirror.
+                                      # Will have to capture manually.
     };
 
 };
@@ -187,7 +199,8 @@ rule MciFir
 
 group gMciFir attntype CHECK_STOP, RECOVERABLE, SPECIAL, UNIT_CS
       filter priority(  8, 9, 22, 23, 6, 0, 40, # Channel failure
-                       20, 5, 10 )              # Recoverable
+                       20, 5, 10 ),             # Recoverable
+             secondarybits( 32 )
 {
     /** MCIFIR[0]
      *  MCIFIRQ_REPLAY_TIMEOUT
@@ -361,9 +374,9 @@ group gMciFir attntype CHECK_STOP, RECOVERABLE, SPECIAL, UNIT_CS
     (MciFir, bit(31)) ? defaultMaskedError;
 
     /** MCIFIR[32]
-     *  MCIFIRQ_MIRROR_ACTION_OCCURRED
+     *  Mirror action occurred
      */
-    (MciFir, bit(32)) ? defaultMaskedError;
+    (MciFir, bit(32)) ? handleMirrorAction;
 
     /** MCIFIR[33]
      *  MCIFIRQ_NONFOREIGN_ACCESS_TO_FOREIGN_BAR
@@ -568,3 +581,12 @@ actionclass calloutDmiBusSelfLowConnMed
     calloutDmiBusSymFru;
     threshold1;
 };
+
+/** Handles memory mirror action event */
+actionclass handleMirrorAction
+{
+    calloutSelfMed;
+    threshold( field(33 / 30 min) );
+    funccall( "handleMirrorAction" );
+};
+
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C b/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C
index 083fd95fd..9fdea8e85 100755
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfP8Mcs.C
@@ -291,6 +291,85 @@ PRDF_PLUGIN_DEFINE( Mcs, dd1mcifirBit##BITNUM );
 PLUGIN_MCIFIR_DD1_CHECK( 48 )
 PLUGIN_MCIFIR_DD1_CHECK( 49 )
 
+//------------------------------------------------------------------------------
+
+/**
+ * @brief   Handles a memory mirror action event.
+ * @param   i_mcsChip An MCS chip.
+ * @param   i_sc      The step code data struct
+ * @returns SUCCESS always
+ */
+int32_t handleMirrorAction( ExtensibleChip * i_mcsChip,
+                            STEP_CODE_DATA_STRUCT & i_sc )
+{
+    #define PRDF_FUNC "[ClearMbsSecondaryBits] "
+
+    int32_t l_rc = SUCCESS;
+
+    do
+    {
+        // Get the primary MCS of the mirrored pair.
+        P8McsDataBundle * mcsdb = getMcsDataBundle( i_mcsChip );
+        ExtensibleChip * primcs = mcsdb->getPrimaryMirroredMcs();
+        if ( NULL == primcs )
+        {
+            PRDF_ERR( PRDF_FUNC"getPrimaryMirroredMcs() failed: "
+                      "i_mcsChip=0x%08x", i_mcsChip->GetId() );
+            break;
+        }
+
+        // Manually capture the registers needed from the primary MCS.
+        CaptureData & cd = i_sc.service_data->GetCaptureData();
+        primcs->CaptureErrorData( cd, Util::hashString("MirrorRegs") );
+
+        SCAN_COMM_REGISTER_CLASS * reg = primcs->getRegister("MCHWFM");
+        l_rc = reg->Read();
+        if ( SUCCESS != l_rc )
+        {
+            PRDF_ERR( PRDF_FUNC"Read() failed on MCHWFM: primcs=0x%08x",
+                      primcs->GetId() );
+            break;
+        }
+
+        if ( i_sc.service_data->IsAtThreshold() )
+        {
+            if ( reg->IsBitSet(1) ) // Mirror disabled.
+            {
+                // Error log will be predictive.
+                i_sc.service_data->SetErrorSig( PRDFSIG_MirrorActionTH );
+            }
+            else // Mirror still enabled but getting a flood of attentions.
+            {
+                // Just submit the error log as hidden.
+                i_sc.service_data->Nologging();
+                i_sc.service_data->Gard( GardAction::NoGard );
+            }
+        }
+        else // Under threshold
+        {
+            // Re-enable the mirror.
+            reg->ClearBit(1);
+            reg->ClearBit(2);
+            reg->ClearBit(3);
+
+            l_rc = reg->Write();
+            if ( SUCCESS != l_rc )
+            {
+                PRDF_ERR( PRDF_FUNC"Write() failed on MCHWFM: primcs=0x%08x",
+                          primcs->GetId() );
+                break;
+            }
+        }
+
+    } while (0);
+
+    return SUCCESS;
+
+    #undef PRDF_FUNC
+}
+PRDF_PLUGIN_DEFINE( Mcs, handleMirrorAction );
+
+
 } // end namespace Mcs
 } // end namespace PRDF
 
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsDataBundle.H b/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsDataBundle.H
index 053b472e8..338a99468 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsDataBundle.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsDataBundle.H
@@ -5,7 +5,9 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2013,2014              */
+/* Contributors Listed Below - COPYRIGHT 2013,2014                        */
+/* [+] International Business Machines Corp.                              */
+/*                                                                        */
 /*                                                                        */
 /* Licensed under the Apache License, Version 2.0 (the "License");        */
 /* you may not use this file except in compliance with the License.       */
@@ -32,6 +34,7 @@
 #include <prdfExtensibleChip.H>
 #include <prdfGlobal.H>
 #include <prdfPlatServices.H>
+#include <prdfParserEnums.H>
 
 namespace PRDF
 {
@@ -48,7 +51,7 @@ class P8McsDataBundle : public DataBundle
      * @param i_mcsChip The MCS chip.
      */
     explicit P8McsDataBundle( ExtensibleChip * i_mcsChip ) :
-        iv_mcsChip(i_mcsChip), iv_membChip(NULL)
+        iv_mcsChip(i_mcsChip), iv_membChip(NULL), iv_primMirMcsChip(NULL)
     {}
 
     /**
@@ -74,6 +77,43 @@ class P8McsDataBundle : public DataBundle
         return iv_membChip;
     }
 
+    /** @return The primary MCS of a mirrored MCS pair. */
+    ExtensibleChip * getPrimaryMirroredMcs()
+    {
+        using namespace TARGETING;
+        using namespace PlatServices;
+
+        do
+        {
+            if ( NULL != iv_primMirMcsChip ) break; // nothing to do.
+
+            TargetHandle_t mcs = iv_mcsChip->GetChipHandle();
+            uint32_t pos       = getTargetPosition(mcs);
+
+            if ( MAX_MCS_PER_PROC <= pos ) break; // can't continue, return NULL
+
+            if ( 0 == (pos & 0x1) ) // This is the primary.
+            {
+                iv_primMirMcsChip = iv_mcsChip;
+            }
+            else // The other MCS is the primary.
+            {
+                pos &= ~0x1; // get even position
+
+                TargetHandle_t proc = getConnectedParent( mcs, TYPE_PROC );
+                if ( NULL == proc ) break; // can't continue, return NULL
+
+                TargetHandle_t pmcs = getConnectedChild( proc, TYPE_MCS, pos );
+                if ( NULL == pmcs ) break; // can't continue, return NULL
+
+                iv_primMirMcsChip = (ExtensibleChip *)systemPtr->GetChip(pmcs);
+            }
+
+        } while (0);
+
+        return iv_primMirMcsChip;
+    }
+
   private: // functions
 
     P8McsDataBundle( const P8McsDataBundle & );
@@ -83,6 +123,9 @@ class P8McsDataBundle : public DataBundle
 
     ExtensibleChip * iv_mcsChip;  ///< This MCS chip
     ExtensibleChip * iv_membChip; ///< The connected MEMBUF chip
+
+    /** The primary MCS of a mirrored MCS pair. */
+    ExtensibleChip * iv_primMirMcsChip;
 };
 
 //------------------------------------------------------------------------------
diff --git a/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsExtraSig.H b/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsExtraSig.H
index 5c366480d..ff061ddc0 100644
--- a/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsExtraSig.H
+++ b/src/usr/diag/prdf/common/plat/pegasus/prdfP8McsExtraSig.H
@@ -5,7 +5,9 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2014                   */
+/* Contributors Listed Below - COPYRIGHT 2014                             */
+/* [+] International Business Machines Corp.                              */
+/*                                                                        */
 /*                                                                        */
 /* Licensed under the Apache License, Version 2.0 (the "License");        */
 /* you may not use this file except in compliance with the License.       */
@@ -31,4 +33,7 @@ PRDR_ERROR_SIGNATURE(MciFir_48_DD1Signature, 0x55550000, "MCIFIR[48]",
 PRDR_ERROR_SIGNATURE(MciFir_49_DD1Signature, 0x55550001, "MCIFIR[49]",
                                       "DD1: MCIFIRQ_INTERNAL_SCOM_ERROR_CLONE");
 
+PRDR_ERROR_SIGNATURE(MirrorActionTH, 0x55550010, "MCIFIR[32]",
+                     "Mirror disabled due to too many UEs");
+
 #endif // __prdfP8McsExtraSig_H