MDIA: add hcdb/HWAS_CHANGED_BIT_MEMDIAG support

  Also add check for the new ATTR_RUN_MAX_MEM_PATTERNS

Change-Id: I343f1054cf657e06cea688f0a87340bdd32f01f2
RTC: 38367
Reviewed-on: http://gfw160.austin.ibm.com:8080/gerrit/5548
Tested-by: Jenkins Server
Reviewed-by: Zane Shelley <zshelle@us.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

7 files changed, 280 insertions, 24 deletions

diff --git a/src/usr/diag/mdia/mdia.C b/src/usr/diag/mdia/mdia.C
index c14bec331..cafc276fb 100644
--- a/src/usr/diag/mdia/mdia.C
+++ b/src/usr/diag/mdia/mdia.C
@@ -56,6 +56,17 @@ errlHndl_t runStep(const TargetHandleList & i_targetList)
     if(top)
     {
         globals.mfgPolicy = top->getAttr<ATTR_MNFG_FLAGS>();
+
+        uint8_t maxMemPatterns =
+            top->getAttr<ATTR_RUN_MAX_MEM_PATTERNS>();
+
+        // This registry / attr is the same as the
+        // exhaustive mnfg one
+        if(maxMemPatterns)
+        {
+            globals.mfgPolicy |=
+              MNFG_FLAG_BIT_MNFG_ENABLE_EXHAUSTIVE_PATTERN_TEST;
+        }
     }
 
     // get the workflow for each target mba passed in.
diff --git a/src/usr/diag/mdia/mdiafwd.H b/src/usr/diag/mdia/mdiafwd.H
index 83186f23e..6737cb29b 100644
--- a/src/usr/diag/mdia/mdiafwd.H
+++ b/src/usr/diag/mdia/mdiafwd.H
@@ -31,6 +31,7 @@
 #include <mdia/mdia.H>
 #include <mdia/mdiareasoncodes.H>
 #include <targeting/common/attributes.H>
+#include <targeting/common/targetservice.H>
 #include <vector>
 #include <map>
 
@@ -184,5 +185,24 @@ errlHndl_t getMbaWorkFlow(
  */
 void doStepCleanup(const Globals & i_globals);
 
+/**
+ * @brief check if hw state has been changed for an mba's
+ *        associated targets
+ *
+ * @param[in] i_mba input mba target
+ *
+ * @retval true if hw state has been changed or else
+ */
+bool isHWStateChanged(TARGETING::TargetHandle_t i_mba);
+
+/**
+ * @brief clear hw changed state for an mba's
+ *        associated targets
+ *
+ * @param[in] i_mba input mba target
+ *
+ */
+void clearHWStateChanged(TARGETING::TargetHandle_t i_mba);
+
 }
 #endif
diff --git a/src/usr/diag/mdia/mdiamba.C b/src/usr/diag/mdia/mdiamba.C
index fdac9f11d..35abedbf6 100644
--- a/src/usr/diag/mdia/mdiamba.C
+++ b/src/usr/diag/mdia/mdiamba.C
@@ -28,6 +28,8 @@
 #include "mdiafwd.H"
 #include "mdiaglobals.H"
 #include "mdiasm.H"
+#include "mdiatrace.H"
+#include "targeting/common/utilFilter.H"
 
 using namespace TARGETING;
 
@@ -59,6 +61,20 @@ errlHndl_t getMbaDiagnosticMode(
         o_mode = ONE_PATTERN;
     }
 
+    // Only need to check hw changed state attributes
+    // when not already set to standard or exhaustive
+    if((FOUR_PATTERNS != o_mode) ||
+       (NINE_PATTERNS != o_mode))
+    {
+        if(isHWStateChanged(i_mba))
+        {
+            o_mode = FOUR_PATTERNS;
+        }
+    }
+
+    MDIA_FAST("getMbaDiagnosticMode: mba: %x, o_mode: 0x%x",
+              get_huid(i_mba), o_mode);
+
     return 0;
 }
 
@@ -119,6 +135,184 @@ errlHndl_t getMbaWorkFlow(DiagMode i_mode, WorkFlow & o_wf)
             break;
     }
 
+    // clear HW changed state attribute
+    o_wf.push_back(CLEAR_HW_CHANGED_STATE);
+
     return 0;
 }
+
+/*
+ *  Local helper function to return a list of Centaur
+ *  DIMMs, and MCS associated with the input MBA target
+ *
+ *  If i_queryOnly = true (Query)
+ *    - Return a list of DIMMs, Centaur, and
+ *      MCS connected to this MBA
+ *
+ *  Else (Clear)
+ *   - Return a list of DIMMs and
+ *     (Centaur + MCS) if all the DIMMs behind this
+ *     Centaur have hwchangedState flags cleared
+ *     or about to be cleared by this MBA
+ */
+TargetHandleList getMemTargetsForQueryOrClear(
+                    TargetHandle_t i_mba, bool i_queryOnly)
+{
+    #define FUNC "getMemTargetsForQueryOrClear: "
+    TargetHandleList o_list;
+
+    do
+    {
+        // add associated DIMMs
+        TargetHandleList dimmList;
+        getChildAffinityTargets(dimmList,
+                                i_mba,
+                                CLASS_NA,
+                                TYPE_DIMM);
+
+        if( ! dimmList.empty() )
+        {
+            o_list.insert(o_list.begin(), dimmList.begin(),
+                          dimmList.end());
+        }
+
+        // add associated Centaur
+        TargetHandleList targetList;
+        getParentAffinityTargets(targetList,
+                                 i_mba,
+                                 CLASS_CHIP,
+                                 TYPE_MEMBUF);
+
+        if( targetList.empty() )
+        {
+            MDIA_FAST(FUNC"no connected centaur "
+                    "for mba: %x", get_huid(i_mba));
+            break;
+        }
+
+        TargetHandle_t centaur = targetList[0];
+
+        // if query flag is not set, check to make sure
+        // all of the dimms connected to this centaur
+        // have cleared hw chagned state attributes
+        // before adding this centaur/mcs to the list.
+        // This is needed because we only clear
+        // the centaur/mcs attribute when all of the
+        // dimms' attributes from both mbas have cleared.
+        if(false == i_queryOnly)
+        {
+            targetList.clear();
+            getChildAffinityTargets(targetList,
+                                    centaur,
+                                    CLASS_NA,
+                                    TYPE_DIMM);
+
+            if( ! targetList.empty() )
+            {
+                TargetHandleList::iterator target;
+
+                for(target = targetList.begin();
+                    target != targetList.end(); ++target)
+                {
+                    // exclude dimms belong to the current mba
+                    // because their attributes will be cleared
+                    if(dimmList.end() !=
+                             std::find(dimmList.begin(),
+                                       dimmList.end(), *target))
+                    {
+                        continue;
+                    }
+
+                    ATTR_HWAS_STATE_CHANGED_FLAG_type hwChangeFlag;
+                    hwChangeFlag =
+                      (*target)->getAttr<ATTR_HWAS_STATE_CHANGED_FLAG>();
+
+                    if(HWAS_CHANGED_BIT_MEMDIAG & hwChangeFlag)
+                    {
+                        MDIA_FAST(FUNC"hwChangedState is not cleared "
+                                  "for dimm: %x", get_huid(*target));
+                        centaur = NULL; // don't add centaur and mcs
+                        break;
+                    }
+                }
+            }
+        }
+
+        if(NULL == centaur)
+        {
+            break;
+        }
+
+        o_list.push_back(centaur);
+
+        // get connected mcs target
+        targetList.clear();
+
+        getParentAffinityTargets(targetList,
+                                 centaur,
+                                 CLASS_UNIT,
+                                 TYPE_MCS);
+
+        if( ! targetList.empty() )
+        {
+            o_list.push_back(targetList[0]);
+        }
+
+    } while(0);
+
+    MDIA_DBG(FUNC"mba: %x, size: %d",
+             get_huid(i_mba), o_list.size());
+
+    return o_list;
+
+    #undef FUNC
+}
+
+
+bool isHWStateChanged(TargetHandle_t i_mba)
+{
+    bool hwChanged = false;
+    ATTR_HWAS_STATE_CHANGED_FLAG_type hwChangeFlag;
+
+    // Get a list of associated targets for attribute query
+    TargetHandleList targetList =
+        getMemTargetsForQueryOrClear(i_mba, true);
+
+    for(TargetHandleList::iterator target = targetList.begin();
+        target != targetList.end(); ++target )
+    {
+        hwChangeFlag =
+            (*target)->getAttr<ATTR_HWAS_STATE_CHANGED_FLAG>();
+
+        if(HWAS_CHANGED_BIT_MEMDIAG & hwChangeFlag)
+        {
+            MDIA_DBG("isHWStateChanged: set for target: %x",
+                     get_huid(*target));
+            hwChanged = true;
+            break;
+        }
+    }
+
+    return hwChanged;
+}
+
+void clearHWStateChanged(TargetHandle_t i_mba)
+{
+    TargetHandleList targetList;
+
+    // Get a list of associated targets for attribute clearing
+    targetList = getMemTargetsForQueryOrClear(i_mba, false);
+
+    for(TargetHandleList::iterator target = targetList.begin();
+        target != targetList.end(); ++target)
+    {
+        MDIA_DBG("clearHWStateChanged: mba: %x, target: %x",
+                 get_huid(i_mba), get_huid(*target));
+
+        clear_hwas_changed_bit( *target,
+                                HWAS_CHANGED_BIT_MEMDIAG);
+    }
+}
+
+
 }
diff --git a/src/usr/diag/mdia/mdiasm.C b/src/usr/diag/mdia/mdiasm.C
index 7a6c542ff..ec2892ba8 100644
--- a/src/usr/diag/mdia/mdiasm.C
+++ b/src/usr/diag/mdia/mdiasm.C
@@ -480,6 +480,7 @@ bool StateMachine::workItemIsAsync(WorkFlowProperties & i_wfp)
     {
         case RESTORE_DRAM_REPAIRS:
         case DUMMY_SYNC_PHASE:
+        case CLEAR_HW_CHANGED_STATE:
 
             // no attention associated with these so
             // schedule the next work item now
@@ -545,6 +546,16 @@ bool StateMachine::executeWorkItem(WorkFlowProperties * i_wfp)
 
                 break;
 
+            case CLEAR_HW_CHANGED_STATE:
+
+                mutex_lock(&iv_mutex);
+
+                clearHWStateChanged(getTarget(*i_wfp));
+
+                mutex_unlock(&iv_mutex);
+
+                break;
+
             default:
                 break;
         }
@@ -822,6 +833,7 @@ CommandMonitor & StateMachine::getMonitor()
 
 bool StateMachine::processMaintCommandEvent(const MaintCommandEvent & i_event)
 {
+
     enum
     {
         CLEANUP_CMD = 0x8,
@@ -870,8 +882,8 @@ bool StateMachine::processMaintCommandEvent(const MaintCommandEvent & i_event)
 
         cmd = static_cast<mss_MaintCmd *>(wfp.data);
 
-        MDIA_FAST("sm: processing event for: %x: cmd: %p",
-                get_huid(getTarget(wfp)));
+        MDIA_FAST("sm: processing event for: %x, cmd: %p, type: %x",
+                get_huid(getTarget(wfp)), cmd, i_event.type);
 
         switch(i_event.type)
         {
diff --git a/src/usr/diag/mdia/mdiasm.H b/src/usr/diag/mdia/mdiasm.H
index 6dccaa86c..f4c93b699 100644
--- a/src/usr/diag/mdia/mdiasm.H
+++ b/src/usr/diag/mdia/mdiasm.H
@@ -55,6 +55,7 @@ enum WorkFlowPhase
     DUMMY_ASYNC_PHASE,
     RESTORE_DRAM_REPAIRS,
     START_SCRUB,
+    CLEAR_HW_CHANGED_STATE,
 };
 
 /**
diff --git a/src/usr/diag/mdia/test/mdiatestmba.H b/src/usr/diag/mdia/test/mdiatestmba.H
index b855c66a2..3dd2d8bd9 100644
--- a/src/usr/diag/mdia/test/mdiatestmba.H
+++ b/src/usr/diag/mdia/test/mdiatestmba.H
@@ -31,6 +31,7 @@
 #include <builtins.h>
 #include <cxxtest/TestSuite.H>
 #include <targeting/common/target.H>
+#include <targeting/common/utilFilter.H>
 #include "../mdiafwd.H"
 #include "../mdiaglobals.H"
 #include "../mdiasm.H"
@@ -46,21 +47,28 @@ class MdiaMbaTest : public CxxTest::TestSuite
 
             TS_TRACE(ENTER_MRK "testGetMbaDiagnosticMode");
 
-            DiagMode mode;
-            Globals globals = {};
-            TargetHandle_t mba = 0;
+            TargetHandleList mbaList;
+            getAllChiplets(mbaList, TYPE_MBA);
 
-            errlHndl_t err = getMbaDiagnosticMode(
-                    globals, mba, mode);
-
-            if(err)
+            if( !mbaList.empty() )
             {
-                TS_FAIL("getMbaDiagnosticMode failed unexpectedly");
-            }
+                DiagMode mode;
+                Globals globals = {};
+                TargetHandle_t mba = mbaList[0];
 
-            if(mode != INIT_ONLY)
-            {
-                TS_FAIL("mode != INIT_ONLY");
+                errlHndl_t err = getMbaDiagnosticMode(
+                                        globals, mba, mode);
+
+                if(err)
+                {
+                    TS_FAIL("getMbaDiagnosticMode failed "
+                            "unexpectedly");
+                }
+
+                if(mode != INIT_ONLY)
+                {
+                    TS_FAIL("mode != INIT_ONLY");
+                }
             }
 
             TS_TRACE(EXIT_MRK "testGetMbaDiagnosticMode");
@@ -76,24 +84,33 @@ class MdiaMbaTest : public CxxTest::TestSuite
             Globals globals = {};
             TargetHandle_t mba = 0;
             DiagMode mode;
+            errlHndl_t err = NULL;
+            TargetHandleList mbaList;
+            getAllChiplets(mbaList, TYPE_MBA);
 
-            errlHndl_t err = getMbaDiagnosticMode(
-                    globals, mba, mode);
-
-            if(err)
+            if( !mbaList.empty() )
             {
-                TS_FAIL("getMbaDiagnosticMode failed unexpectedly");
-            }
+                mba = mbaList[0];
+                err = getMbaDiagnosticMode(
+                                globals, mba, mode);
 
-            if(mode != INIT_ONLY)
-            {
-                TS_FAIL("mode != INIT_ONLY");
+                if(err)
+                {
+                    TS_FAIL("getMbaDiagnosticMode "
+                            "failed unexpectedly");
+                }
+
+                if(mode != INIT_ONLY)
+                {
+                    TS_FAIL("mode != INIT_ONLY");
+                }
             }
 
             WorkFlow wf, expected;
 
             expected.push_back(RESTORE_DRAM_REPAIRS);
             expected.push_back(START_PATTERN_0);
+            expected.push_back(CLEAR_HW_CHANGED_STATE);
 
             err = getMbaWorkFlow(mode, wf);
 
diff --git a/src/usr/diag/mdia/test/mdiatestsm.H b/src/usr/diag/mdia/test/mdiatestsm.H
index 11e675029..fd905f9f0 100644
--- a/src/usr/diag/mdia/test/mdiatestsm.H
+++ b/src/usr/diag/mdia/test/mdiatestsm.H
@@ -311,7 +311,8 @@ class MdiaSmTest : public CxxTest::TestSuite
                 Checks(START_PATTERN_5, true),
                 Checks(START_PATTERN_6, true),
                 Checks(START_PATTERN_7, true),
-                Checks(START_SCRUB, true)
+                Checks(START_SCRUB, true),
+                Checks(CLEAR_HW_CHANGED_STATE, false)
             };
 
             StateMachine sm;