PRD: scrub resume counter for MBA runtime scrub commands

Change-Id: I27153ed86a3db5ab2477d8bdd9fa9b560e8d31f6
RTC: 191647
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/58810
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Matt Derksen <mderkse1@us.ibm.com>
Reviewed-by: Caleb N. Palmer <cnpalmer@us.ibm.com>
Reviewed-by: Benjamin J. Weisenbeck <bweisenb@us.ibm.com>
Reviewed-by: Brian J. Stegmiller <bjs@us.ibm.com>
Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/59011
Tested-by: FSP CI Jenkins <fsp-CI-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>

4 files changed, 165 insertions, 85 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemScrubUtils.H b/src/usr/diag/prdf/plat/mem/prdfMemScrubUtils.H
index 3fbc4a895..0dc884f46 100644
--- a/src/usr/diag/prdf/plat/mem/prdfMemScrubUtils.H
+++ b/src/usr/diag/prdf/plat/mem/prdfMemScrubUtils.H
@@ -37,6 +37,18 @@
 namespace PRDF
 {
 
+/** Simple container to keep track of the number of time a scrub command has
+ *  been resumed. */
+class ScrubResumeCounter
+{
+  public:
+    void reset()      { iv_count = 0;                      }
+    void inc()        { if ( 255 != iv_count ) iv_count++; }
+    bool atTh() const { return 16 <= iv_count;             }
+  private:
+    uint8_t iv_count = 0;
+};
+
 /**
  * @brief  Clears the command complete and WAT workaround attentions.
  * @param  i_chip MBA, MCA, or MCBIST.
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
index fc6ca377b..c1a09b62e 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -38,6 +38,7 @@
 #include <prdfTrace.H>
 #include <prdfAssert.h>
 
+#include <prdfCenMbaDataBundle.H>
 #include <prdfMemScrubUtils.H>
 
 #include <iipServiceDataCollector.h>
@@ -664,12 +665,80 @@ uint32_t startTdScrub<TYPE_MCA>( ExtensibleChip * i_chip,
     #undef PRDF_FUNC
 }
 
-//------------------------------------------------------------------------------
-
 //##############################################################################
 //##                   Centaur Maintenance Command wrappers
 //##############################################################################
 
+template<TARGETING::TYPE T>
+uint32_t __startScrub( ExtensibleChip * i_chip, const MemRank & i_rank,
+                       AddrRangeType i_rangeType, uint32_t i_stopCond,
+                       mss_MaintCmd::TimeBaseSpeed i_cmdSpeed );
+
+template<>
+uint32_t __startScrub<TYPE_MBA>( ExtensibleChip * i_chip,
+                                 const MemRank & i_rank,
+                                 AddrRangeType i_rangeType,
+                                 uint32_t i_stopCond,
+                                 mss_MaintCmd::TimeBaseSpeed i_cmdSpeed )
+{
+    #define PRDF_FUNC "[PlatServices::__startScrub<TYPE_MBA>] "
+
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
+
+    uint32_t o_rc = SUCCESS;
+
+    fapi2::Target<fapi2::TARGET_TYPE_MBA> fapiTrgt ( i_chip->getTrgt() );
+    errlHndl_t errl = nullptr;
+
+    do
+    {
+        #ifdef __HOSTBOOT_RUNTIME
+        // Starting a new command. So clear the resume counter.
+        getMbaDataBundle(i_chip)->iv_scrubResumeCounter.reset();
+        #endif
+
+        // Get the address range of the given rank.
+        fapi2::buffer<uint64_t> saddr, eaddr;
+        o_rc = getMemAddrRange<TYPE_MBA>( i_chip, i_rank, saddr, eaddr,
+                                          i_rangeType );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "getMemAddrRange(0x%08x,0x%2x) failed",
+                      i_chip->getHuid(), i_rank.getKey() );
+            break;
+        }
+
+        // Clear all of the counters and maintenance ECC attentions.
+        o_rc = prepareNextCmd<TYPE_MBA>( i_chip );
+        if ( SUCCESS != o_rc )
+        {
+            PRDF_ERR( PRDF_FUNC "prepareNextCmd(0x%08x) failed",
+                      i_chip->getHuid() );
+            break;
+        }
+
+        // Start the scrub command.
+        mss_TimeBaseScrub cmd { fapiTrgt, saddr, eaddr, i_cmdSpeed,
+                                i_stopCond, false };
+        FAPI_INVOKE_HWP( errl, cmd.setupAndExecuteCmd );
+        if ( nullptr != errl )
+        {
+            PRDF_ERR( PRDF_FUNC "setupAndExecuteCmd() on 0x%08x,0x%02x failed",
+                      i_chip->getHuid(), i_rank.getKey() );
+            PRDF_COMMIT_ERRL( errl, ERRL_ACTION_REPORT );
+            o_rc = FAIL; break;
+        }
+
+    } while (0);
+
+    return o_rc;
+
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
 template<>
 uint32_t startBgScrub<TYPE_MBA>( ExtensibleChip * i_chip,
                                  const MemRank & i_rank )
@@ -708,17 +777,6 @@ uint32_t startBgScrub<TYPE_MBA>( ExtensibleChip * i_chip,
                                             : mss_MaintCmd::BG_SCRUB;
     do
     {
-        // Get the first address of the given rank.
-        fapi2::buffer<uint64_t> saddr, eaddr;
-        o_rc = getMemAddrRange<TYPE_MBA>( i_chip, i_rank, saddr, eaddr,
-                                          SLAVE_RANK );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "getMemAddrRange(0x%08x,0x%2x) failed",
-                      i_chip->getHuid(), i_rank.getKey() );
-            break;
-        }
-
         // Set the required thresholds for background scrubbing.
         o_rc = setBgScrubThresholds<TYPE_MBA>( i_chip, i_rank );
         if ( SUCCESS != o_rc )
@@ -728,27 +786,8 @@ uint32_t startBgScrub<TYPE_MBA>( ExtensibleChip * i_chip,
             break;
         }
 
-        // Clear all of the counters and maintenance ECC attentions.
-        o_rc = prepareNextCmd<TYPE_MBA>( i_chip );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "prepareNextCmd(0x%08x) failed",
-                      i_chip->getHuid() );
-            break;
-        }
-
-        // Start the background scrub command.
-        mss_TimeBaseScrub cmd { fapiTrgt, saddr, eaddr, cmdSpeed,
-                                stopCond, false };
-        errlHndl_t errl = nullptr;
-        FAPI_INVOKE_HWP( errl, cmd.setupAndExecuteCmd );
-        if ( nullptr != errl )
-        {
-            PRDF_ERR( PRDF_FUNC "setupAndExecuteCmd() on 0x%08x,0x%02x failed",
-                      i_chip->getHuid(), i_rank.getKey() );
-            PRDF_COMMIT_ERRL( errl, ERRL_ACTION_REPORT );
-            o_rc = FAIL; break;
-        }
+        o_rc = __startScrub<TYPE_MBA>( i_chip, i_rank, SLAVE_RANK, stopCond,
+                                       cmdSpeed );
 
     } while (0);
 
@@ -774,8 +813,6 @@ uint32_t startTdScrub<TYPE_MBA>( ExtensibleChip * i_chip,
     PRDF_ASSERT( nullptr != i_chip );
     PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
 
-    uint32_t o_rc = SUCCESS;
-
     // Make sure there is a command complete attention when the command stops.
     i_stopCond |= mss_MaintCmd::ENABLE_CMD_COMPLETE_ATTENTION;
 
@@ -811,47 +848,8 @@ uint32_t startTdScrub<TYPE_MBA>( ExtensibleChip * i_chip,
 
     #endif
 
-    do
-    {
-        // Get the address range of the given rank.
-        fapi2::buffer<uint64_t> saddr, eaddr;
-        o_rc = getMemAddrRange<TYPE_MBA>( i_chip, i_rank, saddr, eaddr,
-                                          i_rangeType );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "getMemAddrRange(0x%08x,0x%2x) failed",
-                      i_chip->getHuid(), i_rank.getKey() );
-            break;
-        }
-
-        // Clear all of the counters and maintenance ECC attentions.
-        o_rc = prepareNextCmd<TYPE_MBA>( i_chip );
-        if ( SUCCESS != o_rc )
-        {
-            PRDF_ERR( PRDF_FUNC "prepareNextCmd(0x%08x) failed",
-                      i_chip->getHuid() );
-            break;
-        }
-
-        // Get the MBA fapi target.
-        fapi2::Target<fapi2::TARGET_TYPE_MBA> fapiTrgt ( i_chip->getTrgt() );
-
-        // Start the background scrub command.
-        mss_TimeBaseScrub cmd { fapiTrgt, saddr, eaddr, cmdSpeed,
-                                i_stopCond, false };
-        errlHndl_t errl = nullptr;
-        FAPI_INVOKE_HWP( errl, cmd.setupAndExecuteCmd );
-        if ( nullptr != errl )
-        {
-            PRDF_ERR( PRDF_FUNC "setupAndExecuteCmd() on 0x%08x,0x%02x failed",
-                      i_chip->getHuid(), i_rank.getKey() );
-            PRDF_COMMIT_ERRL( errl, ERRL_ACTION_REPORT );
-            o_rc = FAIL; break;
-        }
-
-    } while (0);
-
-    return o_rc;
+    return __startScrub<TYPE_MBA>( i_chip, i_rank, i_rangeType, i_stopCond,
+                                   cmdSpeed );
 
     #undef PRDF_FUNC
 }
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices_rt.C b/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
index 0a504c5ea..52ca3ef46 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices_rt.C
@@ -36,6 +36,7 @@
 #include <prdfTrace.H>
 
 // Platform includes
+#include <prdfCenMbaDataBundle.H>
 #include <prdfMemScrubUtils.H>
 #include <prdfPlatServices.H>
 
@@ -243,27 +244,85 @@ uint32_t stopBgScrub<TYPE_MBA>( ExtensibleChip * i_chip )
 
 //------------------------------------------------------------------------------
 
+template<TARGETING::TYPE T>
+uint32_t __resumeScrub( ExtensibleChip * i_chip,
+                        AddrRangeType i_rangeType, uint32_t i_stopCond,
+                        mss_MaintCmd::TimeBaseSpeed i_cmdSpeed );
+
 template<>
-uint32_t resumeBgScrub<TYPE_MBA>( ExtensibleChip * i_chip )
+uint32_t __resumeScrub<TYPE_MBA>( ExtensibleChip * i_chip,
+                                  AddrRangeType i_rangeType,
+                                  uint32_t i_stopCond,
+                                  mss_MaintCmd::TimeBaseSpeed i_cmdSpeed )
 {
-    #define PRDF_FUNC "[PlatServices::resumeBgScrub<TYPE_MBA>] "
+    #define PRDF_FUNC "[PlatServices::__resumeScrub<TYPE_MBA>] "
 
     PRDF_ASSERT( nullptr != i_chip );
     PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
 
-    uint32_t rc = SUCCESS;
+    uint32_t o_rc = SUCCESS;
 
-    PRDF_ERR( PRDF_FUNC "function not implemented yet" );
+    if ( getMbaDataBundle(i_chip)->iv_scrubResumeCounter.atTh() )
+    {
+        // We have resumed scrubbing on this rank too many times. We still want
+        // the scrub to continue to the end of the rank, if possible, but we
+        // need to prevent flooding. So mask off all the CE/UE stop-on-error
+        // conditions. Note that there is only one chip mark per rank so we
+        // don't need to worry about getting flooded with those attentions.
+
+        i_stopCond &= ~mss_MaintCmd::STOP_ON_HARD_NCE_ETE;
+        i_stopCond &= ~mss_MaintCmd::STOP_ON_INT_NCE_ETE;
+        i_stopCond &= ~mss_MaintCmd::STOP_ON_SOFT_NCE_ETE;
+        i_stopCond &= ~mss_MaintCmd::STOP_ON_RETRY_CE_ETE;
+        i_stopCond &= ~mss_MaintCmd::STOP_ON_UE;
+    }
+
+    do
+    {
+        // TODO: Clear ECC counters/FIRs. Increment the current address. Clear
+        //       FIRs again. Start the command from the current address to the
+        //       end of the rank.
 
-    /* TODO: RTC 157888 - Not entirely sure how to do this. Will require a inc
-     *       command followed by a start command. May need the stop conditions
-     *       for the start command. */
+        // Resume successful. So increment the resume counter.
+        getMbaDataBundle(i_chip)->iv_scrubResumeCounter.inc();
 
-    return rc;
+    } while (0);
+
+    return o_rc;
 
     #undef PRDF_FUNC
 }
 
+//------------------------------------------------------------------------------
+
+template<>
+uint32_t resumeBgScrub<TYPE_MBA>( ExtensibleChip * i_chip )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
+
+    /* TODO:
+    return __resumeScrub<TYPE_MBA>( i_chip, SLAVE_RANK, stopCond, cmdSpeed );
+    */
+    return SUCCESS;
+}
+
+//------------------------------------------------------------------------------
+
+template<>
+uint32_t resumeTdScrub<TYPE_MBA>( ExtensibleChip * i_chip,
+                                  AddrRangeType i_rangeType,
+                                  uint32_t i_stopCond )
+{
+    PRDF_ASSERT( nullptr != i_chip );
+    PRDF_ASSERT( TYPE_MBA == i_chip->getType() );
+
+    /* TODO:
+    return __resumeScrub<TYPE_MBA>( i_chip, i_rangeType, i_stopCond, cmdSpeed );
+    */
+    return SUCCESS;
+}
+
 //##############################################################################
 //##                       Line Delete Functions
 //##############################################################################
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices_rt.H b/src/usr/diag/prdf/plat/prdfPlatServices_rt.H
index 90b24b2c2..df7cf3156 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices_rt.H
+++ b/src/usr/diag/prdf/plat/prdfPlatServices_rt.H
@@ -95,6 +95,17 @@ uint32_t stopBgScrub( ExtensibleChip * i_chip );
 template<TARGETING::TYPE T>
 uint32_t resumeBgScrub( ExtensibleChip * i_chip );
 
+/**
+ * @brief  Resumes TD scrubbing after it has paused on error.
+ * @param  i_chip      MBA chip.
+ * @param  i_rangeType See enum AddrRangeType.
+ * @param  i_stopCond  The stop conditions for the targeted scrub.
+ * @return Non-SUCCESS if an internal function fails, SUCCESS otherwise.
+ */
+template<TARGETING::TYPE T, typename SCT>
+uint32_t resumeTdScrub( ExtensibleChip * i_chip, AddrRangeType i_rangeType,
+                        SCT i_stopCond );
+
 //##############################################################################
 //##                       Line delete functions
 //##############################################################################