PRD: Axone UE/CE threshold for background scrub

Change-Id: I2872ee94a385d84b10a6e0aaf7f3c0a95c496aa0
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/84551
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Benjamen G Tyner <ben.tyner@ibm.com>
Reviewed-by: Brian J Stegmiller <bjs@us.ibm.com>
Reviewed-by: Zane C Shelley <zshelle@us.ibm.com>
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/84781
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>

4 files changed, 100 insertions, 20 deletions

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfOcmbDataBundle.H b/src/usr/diag/prdf/common/plat/mem/prdfOcmbDataBundle.H
index e5977ee6c..499baf00c 100644
--- a/src/usr/diag/prdf/common/plat/mem/prdfOcmbDataBundle.H
+++ b/src/usr/diag/prdf/common/plat/mem/prdfOcmbDataBundle.H
@@ -40,6 +40,7 @@
 
 #ifdef __HOSTBOOT_MODULE
 
+#include <prdfMemScrubUtils.H>
 #include <prdfMemTdFalseAlarm.H>
 #include <prdfMemThresholds.H>
 #include <prdfMemTdCtlr.H>
@@ -206,6 +207,12 @@ class OcmbDataBundle : public DataBundle
      *  scrubbing is resumed. */
     bool iv_maskMainlineNceTce = false;
 
+    // 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;
+
     #else // IPL only
 
     /** MNFG IPL CE statistics. */
diff --git a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
index 817a345f3..da18cea81 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
+++ b/src/usr/diag/prdf/plat/mem/prdfMemTdCtlr_rt.C
@@ -1693,21 +1693,53 @@ uint32_t MemTdCtlr<TYPE_OCMB_CHIP>::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 ( !getOcmbDataBundle(iv_chip)->iv_ceScrubStopCounter.atTh() &&
+             !getOcmbDataBundle(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 ( !getOcmbDataBundle(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/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
index 491cc344d..55bc84657 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -41,6 +41,7 @@
 
 #include <prdfCenMbaDataBundle.H>
 #include <prdfP9McbistDataBundle.H>
+#include <prdfOcmbDataBundle.H>
 #include <prdfMemScrubUtils.H>
 
 #include <iipServiceDataCollector.h>
@@ -1442,6 +1443,12 @@ uint32_t startBgScrub<TYPE_OCMB_CHIP>( ExtensibleChip * i_ocmb,
     // Get the OCMB fapi target
     fapi2::Target<fapi2::TARGET_TYPE_OCMB_CHIP> fapiTrgt (i_ocmb->getTrgt());
 
+    #ifdef __HOSTBOOT_RUNTIME
+    // Starting a new command. Clear the UE and CE scrub stop counters
+    getOcmbDataBundle( mcbChip )->iv_ueScrubStopCounter.reset();
+    getOcmbDataBundle( 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 cd0b045dc..74ba41e30 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
@@ -38,6 +38,7 @@
 // Platform includes
 #include <prdfCenMbaDataBundle.H>
 #include <prdfP9McbistDataBundle.H>
+#include <prdfOcmbDataBundle.H>
 #include <prdfMemScrubUtils.H>
 #include <prdfPlatServices.H>
 
@@ -467,10 +468,8 @@ uint32_t resumeBgScrub<TYPE_OCMB_CHIP>( ExtensibleChip * i_chip )
     uint32_t o_rc = SUCCESS;
 
     PRDF_TRAC( PRDF_FUNC "Function not supported yet" );
-
-    /* TODO RTC 207273 - no hwp support yet
-
-    // Get the OCMB_CHIP fapi target
+    /* TODO RTC 207273 - no HWP support yet
+    // Get the OCMB fapi target
     fapi2::Target<fapi2::TARGET_TYPE_OCMB_CHIP> fapiTrgt ( i_chip->getTrgt() );
 
     do
@@ -484,9 +483,45 @@ uint32_t resumeBgScrub<TYPE_OCMB_CHIP>( ExtensibleChip * i_chip )
             break;
         }
 
+        // Check UE and CE stop counters to determine stop conditions
+        mss::mcbist::stop_conditions<> stopCond;
+        if ( getOcmbDataBundle(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 ( getOcmbDataBundle(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 )
         {
@@ -497,7 +532,6 @@ uint32_t resumeBgScrub<TYPE_OCMB_CHIP>( ExtensibleChip * i_chip )
         }
 
     } while (0);
-
     */
 
     return o_rc;