PRD: Avoid gard for NVDIMMs

Change-Id: Icaa517b196826c2b442da769ef45b3cdf56e6a9d CQ: SW467502 Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/79189 Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com> Reviewed-by: Brian J. Stegmiller <bjs@us.ibm.com> Reviewed-by: Benjamen G. Tyner <ben.tyner@ibm.com> Reviewed-by: Paul Greenwood <paul.greenwood@ibm.com> Reviewed-by: Zane C. Shelley <zshelle@us.ibm.com> Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/79670
open-power · Jun 28, 2019 · 8db1ba5 · 8db1ba5
1 parent 1581c67
commit 8db1ba5
Show file tree

Hide file tree

Showing 14 changed files with 202 additions and 22 deletions.
diff --git a/src/usr/diag/prdf/common/framework/service/iipServiceDataCollector.h b/src/usr/diag/prdf/common/framework/service/iipServiceDataCollector.h
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2012,2018                        */
+/* Contributors Listed Below - COPYRIGHT 2012,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -627,6 +627,11 @@ class ServiceDataCollector
      */
     void clearMruListGard();
 
+    /**
+     * @brief Iterates the MRU list and clears gard for any NVDIMM targets.
+     */
+    void clearNvdimmMruListGard();
+
     /**
      * @brief  Iterates the MRU list and returns true if at least on target in
      *         the list is set to be garded.

diff --git a/src/usr/diag/prdf/common/framework/service/prdfServiceDataCollector.C b/src/usr/diag/prdf/common/framework/service/prdfServiceDataCollector.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2012,2015                        */
+/* Contributors Listed Below - COPYRIGHT 2012,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -177,6 +177,50 @@ void ServiceDataCollector::clearMruListGard()
 
 //------------------------------------------------------------------------------
 
+void ServiceDataCollector::clearNvdimmMruListGard()
+{
+    #define PRDF_FUNC "[ServiceDataCollector::clearNvdimmMruListGard] "
+
+    // Loop through the MRU list.
+    for ( auto & mru : xMruList )
+    {
+        PRDcallout callout = mru.callout;
+        TargetHandle_t trgt = callout.getTarget();
+        if ( TYPE_DIMM == PlatServices::getTargetType(trgt) )
+        {
+            // If the callout target is an NVDIMM, do not gard it and send a
+            // message to PHYP/Hostboot that a save/restore may work.
+            if ( isNVDIMM(trgt) )
+            {
+                mru.gardState = NO_GARD;
+
+                #ifdef __HOSTBOOT_MODULE
+
+                #ifdef __HOSTBOOT_RUNTIME
+                // Hostboot runtime, send the message to PHYP
+                uint32_t l_rc = PlatServices::nvdimmNotifyPhypProtChange( trgt,
+                    NVDIMM::NVDIMM_RISKY_HW_ERROR );
+                if ( SUCCESS != l_rc )
+                {
+                    PRDF_TRAC( PRDF_FUNC "nvdimmNotifyPhypProtChange(0x%08x) "
+                               "failed.", PlatServices::getHuid(trgt) );
+                    continue;
+                }
+                #else
+                // IPL, set the appropriate internal attribute in Hostboot
+                trgt->setAttr<ATTR_NV_STATUS_FLAG>(0x40);
+                #endif
+
+                #endif // __HOSTBOOT_MODULE
+            }
+        }
+    }
+
+    #undef PRDF_FUNC
+}
+
+//------------------------------------------------------------------------------
+
 bool ServiceDataCollector::isGardRequested()
 {
     bool gardRecordExit = false;

diff --git a/src/usr/diag/prdf/common/plat/mem/prdfMemMark.C b/src/usr/diag/prdf/common/plat/mem/prdfMemMark.C
@@ -1390,6 +1390,9 @@ uint32_t applyRasPolicies( ExtensibleChip * i_chip, const MemRank & i_rank,
     {
         io_sc.service_data->setServiceCall();
 
+        // We want to try to avoid garding NVDIMMs, so clear gard for them now.
+        io_sc.service_data->clearNvdimmMruListGard();
+
         #ifdef __HOSTBOOT_RUNTIME
         // No more repairs left so no point doing any more TPS procedures.
         MemDbUtils::banTps<T>( i_chip, i_rank );

diff --git a/src/usr/diag/prdf/common/plat/nimbus/nimbus_mca.rule b/src/usr/diag/prdf/common/plat/nimbus/nimbus_mca.rule
@@ -241,7 +241,7 @@ group gMCACALFIR
     /** MCACALFIR[0]
      *  A MBA recoverable error has occurred.
      */
-    (rMCACALFIR, bit(0)) ? self_th_1;
+    (rMCACALFIR, bit(0)) ? nvdimm_self_th_1;
 
     /** MCACALFIR[1]
      *  MBA Nonrecoverable Error
@@ -251,7 +251,7 @@ group gMCACALFIR
     /** MCACALFIR[2]
      *  Excessive refreshes to a single rank.
      */
-    (rMCACALFIR, bit(2)) ? self_th_32perDay;
+    (rMCACALFIR, bit(2)) ? nvdimm_self_th_32perDay;
 
     /** MCACALFIR[3]
      *  Err detected in the MBA debug WAT logic
@@ -266,7 +266,7 @@ group gMCACALFIR
     /** MCACALFIR[5]
      *  Calibration complete indication xout
      */
-    (rMCACALFIR, bit(5)) ? self_th_32perDay;
+    (rMCACALFIR, bit(5)) ? nvdimm_self_th_32perDay;
 
     /** MCACALFIR[6]
      *  Emergency Throttle
@@ -533,7 +533,7 @@ group gMCAECCFIR
     /** MCAECCFIR[42]
      *  SCOM_PARITY_CLASS_RECOVERABLE
      */
-    (rMCAECCFIR, bit(42)) ? self_th_1;
+    (rMCAECCFIR, bit(42)) ? nvdimm_self_th_1;
 
     /** MCAECCFIR[43]
      *  SCOM_PARITY_CLASS_UNRECOVERABLE
@@ -548,7 +548,7 @@ group gMCAECCFIR
     /** MCAECCFIR[45]
      *  WRITE_RMW_CE
      */
-    (rMCAECCFIR, bit(45)) ? self_th_32perDay;
+    (rMCAECCFIR, bit(45)) ? nvdimm_self_th_32perDay;
 
     /** MCAECCFIR[46]
      *  WRITE_RMW_UE
@@ -686,12 +686,12 @@ group gDDRPHYFIR
     /** DDRPHYFIR[60]
      *  Register PE 4 bit impact
      */
-    (rDDRPHYFIR, bit(60)) ? self_th_1;
+    (rDDRPHYFIR, bit(60)) ? nvdimm_self_th_1;
 
     /** DDRPHYFIR[61]
      *  Register PE 1 bit impact
      */
-    (rDDRPHYFIR, bit(61)) ? self_th_1;
+    (rDDRPHYFIR, bit(61)) ? nvdimm_self_th_1;
 
 };
 

diff --git a/src/usr/diag/prdf/common/plat/nimbus/nimbus_mca_actions.rule b/src/usr/diag/prdf/common/plat/nimbus/nimbus_mca_actions.rule
@@ -70,6 +70,7 @@ actionclass rcd_parity_error
     calloutSelfLowNoGard;                      # Self LOW
     # Thresholding done in plugin
     funccall("RcdParityError");                # Run TPS on TH for all MCA ranks
+    funccall("ClearNvdimmGardState");          # Clear gard for NVDIMMs
 };
 
 /** Handle Mainline IUEs */
@@ -125,19 +126,36 @@ actionclass maintenance_iaue_handling
 /** MCA/UE algroithm, threshold 5 per day */
 actionclass mca_ue_algorithm_th_5perDay
 {
-    calloutSelfMed;
+    try( funccall("CheckForNvdimms"), calloutSelfMed );
     threshold5pday;
     funccall("mcaUeAlgorithm"); # must be called last
 };
 
 /** MCA/UE algroithm, threshold 1 */
 actionclass mca_ue_algorithm_th_1
 {
-    calloutSelfMed;
+    try( funccall("CheckForNvdimms"), calloutSelfMed );
     threshold1;
     funccall("mcaUeAlgorithm"); # must be called last
 };
 
+################################################################################
+#                               NVDIMM callouts                                #
+################################################################################
+
+# Simple callouts that will avoid gard for NVDIMMs
+actionclass nvdimm_self_th_1
+{
+    try( funccall("CheckForNvdimms"), calloutSelfMed );
+    threshold1;
+};
+
+actionclass nvdimm_self_th_32perDay
+{
+    try( funccall("CheckForNvdimms"), calloutSelfMed );
+    threshold32pday;
+};
+
 ################################################################################
 # Analyze groups
 ################################################################################

diff --git a/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcbist.rule b/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcbist.rule
@@ -5,7 +5,7 @@
 #
 # OpenPOWER HostBoot Project
 #
-# Contributors Listed Below - COPYRIGHT 2016,2018
+# Contributors Listed Below - COPYRIGHT 2016,2019
 # [+] International Business Machines Corp.
 #
 #
@@ -599,7 +599,7 @@ group gMCBISTFIR
     /** MCBISTFIR[13]
      *  SCOM_RECOVERABLE_REG_PE
      */
-    (rMCBISTFIR, bit(13)) ? self_th_1;
+    (rMCBISTFIR, bit(13)) ? nvdimm_self_th_1;
 
     /** MCBISTFIR[14]
      *  SCOM_FATAL_REG_PE

diff --git a/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcbist_actions.rule b/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcbist_actions.rule
@@ -5,7 +5,7 @@
 #
 # OpenPOWER HostBoot Project
 #
-# Contributors Listed Below - COPYRIGHT 2016,2018
+# Contributors Listed Below - COPYRIGHT 2016,2019
 # [+] International Business Machines Corp.
 #
 #
@@ -36,6 +36,17 @@ actionclass command_addr_timeout
     funccall("commandAddrTimeout");
 };
 
+################################################################################
+#                               NVDIMM callouts                                #
+################################################################################
+
+# Simple callouts that will avoid gard for NVDIMMs
+actionclass nvdimm_self_th_1
+{
+    try( funccall("CheckForNvdimms"), calloutSelfMed );
+    threshold1;
+};
+
 ###############################################################################
 # Analyze groups
 ###############################################################################

diff --git a/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcs.rule b/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcs.rule
@@ -5,7 +5,7 @@
 #
 # OpenPOWER HostBoot Project
 #
-# Contributors Listed Below - COPYRIGHT 2016,2018
+# Contributors Listed Below - COPYRIGHT 2016,2019
 # [+] International Business Machines Corp.
 #
 #
@@ -148,7 +148,7 @@ group gMCFIR
     /** MCFIR[0]
      *  mc internal recoverable eror
      */
-    (rMCFIR, bit(0)) ? self_th_1;
+    (rMCFIR, bit(0)) ? nvdimm_self_th_1;
 
     /** MCFIR[1]
      *  mc internal non recovervable error

diff --git a/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcs_actions.rule b/src/usr/diag/prdf/common/plat/nimbus/nimbus_mcs_actions.rule
@@ -5,7 +5,7 @@
 #
 # OpenPOWER HostBoot Project
 #
-# Contributors Listed Below - COPYRIGHT 2018
+# Contributors Listed Below - COPYRIGHT 2018,2019
 # [+] International Business Machines Corp.
 #
 #
@@ -23,6 +23,17 @@
 #
 # IBM_PROLOG_END_TAG
 
+################################################################################
+#                               NVDIMM callouts                                #
+################################################################################
+
+# Simple callouts that will avoid gard for NVDIMMs
+actionclass nvdimm_self_th_1
+{
+    try( funccall("CheckForNvdimms"), calloutSelfMed );
+    threshold1;
+};
+
 ################################################################################
 # Analyze groups
 ################################################################################

diff --git a/src/usr/diag/prdf/common/plat/p9/p9_common_actions.rule b/src/usr/diag/prdf/common/plat/p9/p9_common_actions.rule
@@ -279,4 +279,3 @@ actionclass chip_to_chip
     calloutSelfMed;
     threshold1;
 };
-
diff --git a/src/usr/diag/prdf/common/plat/p9/prdfCommonPlugins.C b/src/usr/diag/prdf/common/plat/p9/prdfCommonPlugins.C
@@ -127,6 +127,66 @@ PRDF_PLUGIN_DEFINE_NS(nimbus_proc,  CommonPlugins, ClearServiceCallFlag_mnfgInfo
 PRDF_PLUGIN_DEFINE_NS(cumulus_proc, CommonPlugins, ClearServiceCallFlag_mnfgInfo);
 PRDF_PLUGIN_DEFINE_NS(axone_proc,   CommonPlugins, ClearServiceCallFlag_mnfgInfo);
 
+/**
+ * @brief   Will change the gard state of any NVDIMMs in the callout list to
+ *          NO_GARD.
+ * @param   i_chip The chip.
+ * @param   io_sc  The step code data struct.
+ * @returns SUCCESS
+ */
+int32_t ClearNvdimmGardState( ExtensibleChip * i_chip,
+                              STEP_CODE_DATA_STRUCT & io_sc )
+{
+    #ifdef __HOSTBOOT_MODULE
+
+    // Call the sdc to clear the NVDIMM mru list.
+    io_sc.service_data->clearNvdimmMruListGard();
+
+    #endif
+
+    return SUCCESS;
+}
+PRDF_PLUGIN_DEFINE_NS(nimbus_mcs,    CommonPlugins, ClearNvdimmGardState);
+PRDF_PLUGIN_DEFINE_NS(nimbus_mca,    CommonPlugins, ClearNvdimmGardState);
+PRDF_PLUGIN_DEFINE_NS(nimbus_mcbist, CommonPlugins, ClearNvdimmGardState);
+
+/**
+ * @brief   Will check if any of the DIMMs connected to this chip are NVDIMMs
+ *          and callout self, no gard if there are.
+ * @param   i_chip The chip of the DIMM parent.
+ * @param   io_sc  The step code data struct.
+ * @returns SUCCESS if NVDIMMs found, PRD_SCAN_COMM_REGISTER_ZERO if not.
+ */
+int32_t CheckForNvdimms( ExtensibleChip * i_chip,
+                         STEP_CODE_DATA_STRUCT & io_sc )
+{
+    int32_t rc = PRD_SCAN_COMM_REGISTER_ZERO;
+
+    #ifdef __HOSTBOOT_MODULE
+
+    TargetHandleList dimmList = getConnected( i_chip->getTrgt(), TYPE_DIMM );
+
+    for ( auto & dimm : dimmList )
+    {
+        if ( isNVDIMM(dimm) )
+        {
+            // Callout self, no gard
+            io_sc.service_data->SetCallout(i_chip->getTrgt(), MRU_MED, NO_GARD);
+
+            // No need for other actions, so return SUCCESS
+            rc = SUCCESS;
+            break;
+        }
+    }
+
+    #endif
+
+    return rc;
+}
+PRDF_PLUGIN_DEFINE_NS(nimbus_mcs,    CommonPlugins, CheckForNvdimms);
+PRDF_PLUGIN_DEFINE_NS(nimbus_mca,    CommonPlugins, CheckForNvdimms);
+PRDF_PLUGIN_DEFINE_NS(nimbus_mcbist, CommonPlugins, CheckForNvdimms);
+
 } // namespace CommonPlugins ends
 
 }// namespace PRDF ends

diff --git a/src/usr/diag/prdf/plat/mem/prdfMemDynDealloc.C b/src/usr/diag/prdf/plat/mem/prdfMemDynDealloc.C
@@ -1351,6 +1351,20 @@ int32_t dimmList( TargetHandleList  & i_dimmList )
         sendPredDeallocRequest( ssAddr, seAddr );
         PRDF_TRAC( PRDF_FUNC "Predictive dealloc for start addr: 0x%016llx "
                    "end addr: 0x%016llx", ssAddr, seAddr );
+
+        // If the DIMM is an NVDIMM, send a message to PHYP that a save/restore
+        // may work.
+        if ( isNVDIMM(*it) )
+        {
+            uint32_t l_rc = PlatServices::nvdimmNotifyPhypProtChange( *it,
+                    NVDIMM::NVDIMM_RISKY_HW_ERROR );
+            if ( SUCCESS != l_rc )
+            {
+                PRDF_TRAC( PRDF_FUNC "nvdimmNotifyPhypProtChange(0x%08x) "
+                           "failed.", getHuid(*it) );
+                continue;
+            }
+        }
     }
 
     return o_rc;

diff --git a/src/usr/diag/prdf/plat/mem/prdfP9Mca.C b/src/usr/diag/prdf/plat/mem/prdfP9Mca.C
@@ -782,9 +782,9 @@ int32_t AnalyzeNvdimmHealthStatRegs( ExtensibleChip * i_chip,
             // and make the log predictive.
             io_sc.service_data->SetThresholdMaskId(0);
 
-            // Send persistency lost message to PHYP
+            // Send message to PHYP that save/restore may work
             l_rc = PlatServices::nvdimmNotifyPhypProtChange( dimm,
-                    NVDIMM::UNPROTECTED_BECAUSE_ERROR );
+                    NVDIMM::NVDIMM_RISKY_HW_ERROR );
             if ( SUCCESS != l_rc ) continue;
 
             // Analyze Health Status0 Reg, Health Status1 Reg,