PRD: Add threshold for stopping on UEs/CEs during BgScrub

Change-Id: Iddeec04300631fc57b5c2f4a2eb57302e9f98fe2
CQ: SW476467
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/84071
Reviewed-by: Benjamen G Tyner <ben.tyner@ibm.com>
Reviewed-by: Paul Greenwood <paul.greenwood@ibm.com>
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Zane C Shelley <zshelle@us.ibm.com>
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/84073
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>

4 files changed, 103 insertions, 16 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
index 779d13a73..817a345f3 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
@@ -1623,21 +1623,53 @@ uint32_t MemTdCtlr<TYPE_MCBIST>::canResumeBgScrub( bool & o_canResume )
     // can use the stop conditions, which should be unique for background scrub,
     // to determine if it has been configured.
 
-    SCAN_COMM_REGISTER_CLASS * reg = iv_chip->getRegister( "MBSTR" );
-    o_rc = reg->Read();
-    if ( SUCCESS != o_rc )
-    {
-        PRDF_ERR( PRDF_FUNC "Read() failed on MBSTR: iv_chip=0x%08x",
-                  iv_chip->getHuid() );
-    }
-    else if ( 0xf != reg->GetBitFieldJustified(0,4) && // NCE int TH
-              0xf != reg->GetBitFieldJustified(4,4) && // NCE soft TH
-              0xf != reg->GetBitFieldJustified(8,4) && // NCE hard TH
-              reg->IsBitSet(34)                     && // pause on MPE
-              reg->IsBitSet(35)                     )  // pause on UE
+    do
     {
-        o_canResume = true;
-    }
+        SCAN_COMM_REGISTER_CLASS * reg = iv_chip->getRegister( "MBSTR" );
+        o_rc = reg->Read();
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "Read() failed on MBSTR: iv_chip=0x%08x",
+                    iv_chip->getHuid() );
+            break;
+        }
+        // Note: The stop conditions for background scrubbing can now be
+        // variable depending on whether we have hit threshold for the number
+        // of UEs or CEs that we have stopped on on a rank.
+
+        // If we haven't hit CE or UE threshold, check the CE stop conditions
+        if ( !getMcbistDataBundle(iv_chip)->iv_ceScrubStopCounter.atTh() &&
+             !getMcbistDataBundle(iv_chip)->iv_ueScrubStopCounter.atTh() )
+        {
+            // If the stop conditions aren't set, just break out.
+            if ( !(0xf != reg->GetBitFieldJustified(0,4) && // NCE int TH
+                   0xf != reg->GetBitFieldJustified(4,4) && // NCE soft TH
+                   0xf != reg->GetBitFieldJustified(8,4)) ) // NCE hard TH
+            {
+                break;
+            }
+
+        }
+
+        // If we haven't hit UE threshold yet, check the UE stop condition
+        if ( !getMcbistDataBundle(iv_chip)->iv_ueScrubStopCounter.atTh() )
+        {
+            // If the stop condition isn't set, just break out
+            if ( !reg->IsBitSet(35) ) // pause on UE
+            {
+                break;
+            }
+        }
+
+        // Need to check the stop on mpe stop condition regardless of whether
+        // we hit the UE or CE threshold.
+        if ( reg->IsBitSet(34) ) // pause on MPE
+        {
+            // If we reach here, all the stop conditions are set for background
+            // scrub, so we can resume.
+            o_canResume = true;
+        }
+    }while(0);
 
     return o_rc;
 
diff --git a/src/usr/diag/prdf/plat/mem/prdfP9McbistDataBundle.H b/src/usr/diag/prdf/plat/mem/prdfP9McbistDataBundle.H
index 4a284253a..3883eb936 100644
--- a/src/usr/diag/prdf/plat/mem/prdfP9McbistDataBundle.H
+++ b/src/usr/diag/prdf/plat/mem/prdfP9McbistDataBundle.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2016                             */
+/* Contributors Listed Below - COPYRIGHT 2016,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -81,6 +81,17 @@ class McbistDataBundle : public DataBundle
 
     /** The Targeted Diagnostics controller. */
     MemTdCtlr<TARGETING::TYPE_MCBIST> * iv_tdCtlr = nullptr;
+
+  public: // instance variables
+    #ifdef __HOSTBOOT_RUNTIME
+
+    // These are used to limit the number of times a scrub command will stop
+    // on a UE or CE on a rank. This is to prevent potential flooding of
+    // maintenance UEs or CEs. The threshold will be 16 per rank for each.
+    ScrubResumeCounter iv_ueScrubStopCounter;
+    ScrubResumeCounter iv_ceScrubStopCounter;
+
+    #endif
 };
 
 /**
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
index 48d26f8b8..e4b122ab5 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -40,6 +40,7 @@
 #include <prdfRegisterCache.H>
 
 #include <prdfCenMbaDataBundle.H>
+#include <prdfP9McbistDataBundle.H>
 #include <prdfMemScrubUtils.H>
 
 #include <iipServiceDataCollector.h>
@@ -791,6 +792,12 @@ uint32_t startBgScrub<TYPE_MCA>( ExtensibleChip * i_mcaChip,
     ExtensibleChip * mcbChip = getConnectedParent( i_mcaChip, TYPE_MCBIST );
     fapi2::Target<fapi2::TARGET_TYPE_MCBIST> fapiTrgt ( mcbChip->getTrgt() );
 
+    #ifdef __HOSTBOOT_RUNTIME
+    // Starting a new command. Clear the UE and CE scrub stop counters
+    getMcbistDataBundle( mcbChip )->iv_ueScrubStopCounter.reset();
+    getMcbistDataBundle( mcbChip )->iv_ceScrubStopCounter.reset();
+    #endif
+
     // Get the stop conditions.
     // NOTE: If HBRT_PRD is not configured, we want to use the defaults so that
     //       background scrubbing never stops.
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices_rt.C b/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
index e3bdecb59..cd0b045dc 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
@@ -37,6 +37,7 @@
 
 // Platform includes
 #include <prdfCenMbaDataBundle.H>
+#include <prdfP9McbistDataBundle.H>
 #include <prdfMemScrubUtils.H>
 #include <prdfPlatServices.H>
 
@@ -173,9 +174,45 @@ uint32_t resumeBgScrub<TYPE_MCBIST>( ExtensibleChip * i_chip )
             break;
         }
 
+        // Check UE and CE stop counters to determine stop conditions
+        mss::mcbist::stop_conditions<> stopCond;
+        if ( getMcbistDataBundle(i_chip)->iv_ueScrubStopCounter.atTh() )
+        {
+            // If we've reached the limit of UEs we're allowed to stop on
+            // per rank, only set the stop on mpe stop condition.
+            stopCond.set_pause_on_mpe(mss::ON);
+        }
+        else if ( getMcbistDataBundle(i_chip)->iv_ceScrubStopCounter.atTh() )
+        {
+            // If we've reached the limit of CEs we're allowed to stop on
+            // per rank, set all the normal stop conditions except stop on CE
+            stopCond.set_pause_on_aue(mss::ON);
+
+            #ifdef CONFIG_HBRT_PRD
+
+            stopCond.set_pause_on_mpe(mss::ON)
+                    .set_pause_on_ue(mss::ON);
+
+            // In MNFG mode, stop on RCE_ETE to get an accurate callout for IUEs
+            if ( mfgMode() ) stopCond.set_thresh_rce(1);
+
+            #endif
+        }
+        else
+        {
+            // If we haven't reached threshold on the number of UEs or CEs we
+            // have stopped on, do not change the stop conditions.
+            stopCond = mss::mcbist::stop_conditions<>(
+                mss::mcbist::stop_conditions<>::DONT_CHANGE );
+        }
+
         // Resume the command on the next address.
+        // Note: we have to limit the number of times a command has been stopped
+        // because of a UE/CE. Therefore, we must always resume the command to
+        // the end of the current slave rank so we can reset the UE/CE counts.
         errlHndl_t errl;
-        FAPI_INVOKE_HWP( errl, mss::memdiags::continue_cmd, fapiTrgt );
+        FAPI_INVOKE_HWP( errl, mss::memdiags::continue_cmd, fapiTrgt,
+            mss::mcbist::STOP_AFTER_SLAVE_RANK, stopCond );
 
         if ( nullptr != errl )
         {